Professional Documents
Culture Documents
The titles below are available from booksellers, or from Cambridge University Press at www.cambridge.org/mathematics
Triangulated Categories
Edited by
THORSTEN HOLM
Leibniz Universität Hannover, Germany
PETER JØRGENSEN
University of Newcastle upon Tyne
R A P H A Ë L RO U QU I E R
University of Oxford
cambridge university press
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore,
São Paulo, Delhi, Dubai, Tokyo
Cambridge University Press
The Edinburgh Building, Cambridge CB2 8RU, UK
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
Information on this title: www.cambridge.org/9780521744317
C Cambridge University Press 2010
A catalogue record for this publication is available from the British Library
v
vi Contents
vii
Triangulated categories:
definitions, properties, and examples
thorsten holm and peter jørgensen
1
2 Thorsten Holm and Peter Jørgensen
1. Additive categories
In this first section we shall discuss the fundamental notion of an additive
category and provide some examples. In particular, the category of complexes
over an additive category is introduced which will play a fundamental role in
the sequel.
(A1) For every pair of objects X, Y the set of morphisms HomA (X, Y ) is an
abelian group and the composition of morphisms
Remark 1.2.
(i) A category satisfying (A1) and (A2) is called a preadditive category.
(ii) We recall the notion of coproduct from category theory. Let C be a category
and X, Y objects in C. A coproduct of X and Y in C is an object X ⊕ Y
together with morphisms ιX : X → X ⊕ Y and ιY : Y → X ⊕ Y satisfying
the following universal property: for every object Z in C and morphisms
fX : X → Z and fY : Y → Z there is a unique morphism f : X ⊕ Y → Z
making the following diagram commutative.
Z
fX 6 @ I f
@Y
f
@
X ι - X ⊕ Y ι Y
X Y
Example 1.3.
(i) Let R be a ring and consider R as a category CR with only one object.
The unique morphism set is the underlying abelian group and composition
of morphisms is given by ring multiplication. Then CR satisfies (A1) and
(A2), thus preadditive categories can be seen as generalizations of rings.
But CR is not additive in general; in fact the coproduct of the unique object
with itself would have to be again this object together with fixed ring
elements ι1 , ι2 , and the universal property would mean that for arbitrary
Triangulated categories 3
fn+1 fn fn−1
? ? ?
... -Y -Y -Y - ...
n+1 n n−1
Proof. (A1) Addition of morphisms is defined degreewise, i.e. for two mor-
phisms f = (fn )n∈Z and g = (gn )n∈Z from X to Y their sum is f + g :=
(fn + gn )n∈Z . Using the additive structure of A it is then easy to check that
(A1) holds.
(A2) The zero object in C(A) is the complex (0A , d) where 0A is the zero
object of the additive category A and all differentials are the unique (zero)
morphism on the zero object.
4 Thorsten Holm and Peter Jørgensen
(A3) The coproduct of two complexes X = (Xn , dnX ) and Y = (Yn , dnY ) is
defined degreewise by using the coproduct in the additive category A. More
precisely X ⊕ Y = (Xn ⊕ Yn , dn )n∈Z where the differential is obtained by the
universal property as in the following diagram.
Xn−1 ⊕ Yn−1
ιXn−1 dnX 6 @ I ιY d Y
dn @ n−1 n
@
-
Xn ιXn Xn ⊕ Yn ιYn Yn
Xn−2 ⊕ Yn−2
6 I
@
@ 0
0
dn−1 dn @
@
Xn - Xn ⊕ Yn Yn
ιXn ιYn
it follows that dn−1 ◦ dn = 0. This complex indeed satisfies the properties of a
coproduct in the category of complexes C(A), with morphisms of complexes
ιX = (ιXn )n∈Z : X → X ⊕ Y and ιY = (ιYn )n∈Z : Y → X ⊕ Y . For checking
the universal property let Z be an arbitrary complex and let fX : X → Z
and fY : Y → Z be arbitrary morphisms. The unique morphism of complexes
satisfying fX = f ◦ ιX and fY = f ◦ ιY is f = (fn )n∈Z : X ⊕ Y → Z, where
fn is obtained from the universal property in degree n as in the following
diagram.
Zn
(fX )n f 6 I
@ (fY )n
@
n
@
Xn ι - Xn ⊕ Yn ι Yn
Xn Yn
Remark 1.5. For complexes over A = R-Mod where R is a ring with unit
(and other similar examples) the coproduct of two complexes is more easily
be described on elements as X ⊕ Y = (Xn ⊕ Yn , dn )n∈Z where the differential
is given by dn (xn , yn ) = (dnX (xn ), dnY (yn )) for xn ∈ Xn and yn ∈ Yn , and with
morphisms ιX : X → X ⊕ Y and ιY : Y → X ⊕ Y being the inclusion maps.
The unique morphism of complexes satisfying fX = f ◦ ιX and fY = f ◦ ιY
is then given by fn (xn , yn ) = fX (xn ) + fY (yn ).
Triangulated categories 5
2. Abelian categories
In this section we shall review the fundamental definition of an abelian category,
including the necessary background on the categorical notions of kernels and
cokernels. The prototype example of an abelian category will be the category
R-Mod of modules over a ring R; but we will also see other examples in due
course.
We first recall some notions from category theory. Let A be an additive
category; in particular for every pair of objects X, Y there is a zero morphism,
namely the composition of the unique morphisms X → 0 → Y involving the
zero object of A.
The kernel of a morphism f : X → Y is an object K together with a mor-
phism k : K → X such that
(i) f ◦ k = 0,
(ii) (universal property) for every morphism k : K → X such that f ◦ k = 0,
there is a unique morphism g : K → K making the following diagram
commutative.
Y
A K
f 6
0 A
A0
K A
k
X
YH
H A
H Ig A
@
H
k H @
HH @A
K
Remark 2.2. Suppose that for a morphism f both the coimage and the image
exist. Then we claim that it follows from the universal properties that there is
a natural morphism coim f → im f .
In fact, the image of f is the kernel of c : Y → coker f , hence there is
a morphism k̃ : im f → Y such that c ◦ k̃ = 0 and by the universal property
there exists a unique morphism g̃ : X → im f making the following diagram
commutative.
coker f
AK
c 6
0 A
A0
k̃ im f A
Y
YH
H
H @ I A
g̃ A
f H@ A
H
H @X
H
coker f
AK
c 6
0 A
A0
k̃ im f A
Y H
YH
HH @ I A
A
0 H@ A
HH @
ker f
Then we can consider the following diagram for the universal property of the
coimage
8 Thorsten Holm and Peter Jørgensen
X
A
k
c̃ A
? A g̃
ker f 0- coim fA
H
HH @ A
A
0 HH @ AU
HH R
@
j im f
Example 2.4.
(i) Let R be a ring. The category R-Mod of all R-modules is an abelian
category. In fact, (A5) follows directly from the isomorphism theorem for
R-modules.
However, the subcategory R-mod of finitely generated modules is not
abelian in general since kernels of homomorphisms between finitely gen-
erated modules need not be finitely generated. Indeed we have that R-mod
is an abelian category if and only if R is Noetherian.
In particular, the category of finite-dimensional vector spaces over a
field is abelian, and the category of finitely generated abelian groups is
abelian.
(ii) The subcategory of Ab consisting of free abelian groups is not abelian.
On the other hand, for a prime number p, the abelian p-groups form
an abelian subcategory of Ab (an abelian group is called a p-group if for
every element a we have pk a = 0 for some k).
(iii) For finding examples of additive categories satisfying (A4) but failing to
be abelian, the following observation can be useful. Suppose f : X → Y
is a morphism with ker f = 0 and coker f = 0, i.e. a monomorphism and
an epimorphism. Then the coimage of f is the identity on X, the image
of f is the identity on Y and hence the natural morphism coim f →
im f is just f itself. So in this special case the axiom (A5) states that
a morphism which is a monomorphism and an epimorphism must be
invertible.
Triangulated categories 9
(iv) Explicit examples of additive categories where axiom (A5) fails for the
above reason are the category of topological abelian groups (with contin-
uous group homomorphisms) or the category of Banach complex vector
spaces (with continuous linear maps). In such categories the cokernel of
a morphism f : X → Y is of the form Y / imf where imf is the closure
of the usual set-theoretic image of f . In particular, the natural morphism
coim f → im f is the inclusion of the usual image of f into its closure,
and this is in general not an isomorphism.
hence also the morphism coim f → im f , are obtained degreewise. But since
A is abelian by assumption, we know that for every n the natural morphism
coim fn → im fn in A is indeed an isomorphism. Then, by the introductory
remark, the morphism of complexes (coim fn → im fn )n∈Z is an isomorphism
in C(A).
Example 2.6. We provide an explicit example for the failure of axiom (A4)
in a homotopy category. Consider the abelian category A = Ab of abelian
groups.
Let f : X → Y be the following morphism of complexes of abelian groups,
with non-zero entries in degrees 1 and 0,
... 0 - 0 - Z - 0 ...
id
? ? ? ?
... 0 - Z id- Z - 0 ...
In the category C(Ab) of complexes f is non-zero and has the zero complex
as kernel (cf. the proof of Proposition 2.5). However, f is homotopic to zero
(with the identity as homotopy map), i.e. f = 0 in the homotopy category
K(Ab).
We claim that in the homotopy category f has no kernel. Recall the cate-
gorical definition of the kernel of a morphism f : X → Y from Section 2.
Suppose for a contradiction that our morphism f had a kernel in K(Ab). So
there is a complex . . . → K1 → K0 → K−1 → . . . and a morphism k = k0 :
K0 → Z of abelian groups (in all other degrees the map k has to be zero since X
is concentrated in degree 0). The image of k, being a subgroup of Z, has the form
rZ for some fixed r ∈ Z. Now choose K = X and consider the morphisms
l : K → X given by multiplication with l for any l ∈ Z. Clearly, f ◦ l = 0 in
K(Ab) since f = 0 in K(Ab). According to the universal property of a kernel,
there must exist (unique) morphisms ul : Z → K0 such that k ◦ ul = l up to
homotopy. However, these maps are from K = X to X and this complex is
concentrated in degree 0. Thus there are no non-zero homotopy maps and so
k ◦ ul = l as morphism of abelian groups. But the image of k ◦ ul is contained
in the image of k which is rZ for a fixed r, so k ◦ ul = l can not hold for
arbitrary l ∈ Z, a contradiction.
Hence axiom (A4) fails and therefore the homotopy category K(Ab) is not
an abelian category.
Triangulated categories 11
Remark 3.2. The above version (TR5) of the octahedral axiom is taken from
the book by Kashiwara and Schapira [7, Sec. 1.4]. There are various other
versions appearing in the literature which are equivalent to (TR5), see for
instance A. Neeman’s article [13] or his book [12]; a short treatment can
also be found in A. Hubery’s notes [6] (which are based on the former
references).
We shall only mention two variations here. Mainly a reformulation of the
axiom (TR5) is the following. Note that in (TR5) the given three distinguished
triangles are placed in the first three rows, whereas in (TR5’) below they are
placed in the first two rows and the second column.
Triangulated categories 13
u v l
(TR5’) Given distinguished triangles X → Y → Z → X, Y → Z → X →
vu s
Y and X → Z → Y → X, then there exists a distinguished triangle Z →
v
Y → X → Z making the following diagram commutative and satisfying
(u)s = lv .
u- - Z - X
X Y
idX v idX
? ? ? ?
vu- - Y s - X
X Z
v
? id ?
X
X
- X
? ?
Y - Z
It is not difficult to check that (TR5) and (TR5’) are indeed equivalent; we leave
this verification as an exercise to the reader.
u- - Z - X
X Y
idX v idX
? ? ? ?
vu- - Y - X
X Z
? ? ? ?
0 - X idX- X - 0
? ? ? ?
X u- Y - Z - 2X
14 Thorsten Holm and Peter Jørgensen
f∗ g∗ h∗ f∗ g∗
? ? ? ? ?
Hom(Z , X ) - Hom(Z , Y ) - Hom(Z , Z ) - Hom(Z , X ) - Hom(Z , Y )
Proof. We first show that u is a split monomorphism, i.e. there exists a mor-
phism u such that u ◦ u = idX . We have the following commutative diagram
of distinguished triangles.
16 Thorsten Holm and Peter Jørgensen
u- v- 0-
X Y Z X
id 0 id
? ? ?
id- 0- 0-
X X 0 X
Example 5.2.
(i) Let R be a semisimple ring. Then the module categories R-Mod and R-mod
are semisimple. In particular, the category of vector spaces VecK over a
field K is semisimple.
(ii) The category Ab of abelian groups is not semisimple. For instance, the short
·2 ·1
exact sequence 0 → Z/2Z −→ Z/4Z −→ Z/2Z → 0 does not split.
The following result illustrates that the concepts of abelian and triangulated
categories overlap only slightly.
a distinguished triangle
f v u
X −→ Y −→ V −→ X
where u = 0. Now the triangle splits by Proposition 4.4.
We shall see in the next section that the homotopy category K(A) of com-
plexes over an additive category A is a triangulated category. This, together
with the preceding theorem, will then give a more structural explanation of the
earlier observation that K(Ab) is not abelian in Example 2.6, where we have
used an ad-hoc argument to show that morphisms do not necessarily have a
kernel.
The next step for getting a triangulated structure on the homotopy category
is to find a suitable set of distinguished triangles. To this end, the following
construction of mapping cones is crucial.
Remark 6.4.
(i) There are canonical morphisms in C(A) as follows
and
Example 6.5.
(i) For any complex X consider the zero map f : X → 0 to the zero complex.
Then the mapping cone is M(f ) = X[1]. On the other hand, the mapping
cone of g : 0 → Y is just M(g) = Y itself.
(ii) Let A and B be objects in A and view them as complexes XA and
XB concentrated in degree 0. Any morphism f : A → B in A induces
a morphism of complexes f : XA → XB . Its mapping cone is the
complex
f
. . . → 0 → A −→ B → 0 → . . .
(iii) Let X = (Xn , dnX ) be any complex in C(A). The mapping cone of the
identity morphism idX has degree n term equal to Xn−1 ⊕ Xn and differ-
ential
−dn−1X
0
: Xn−1 ⊕ Xn → Xn−2 ⊕ Xn−1 .
idXn−1 dnX
It is easy to check that the morphisms α(f ) and β(f ) are also well-defined in
the homotopy category K(A) (i.e. independent on the choice of representatives
of the equivalence class of morphisms). This leads to the following definition.
Proof. We have to show that with the above translation functor [1] and the set
of distinguished triangles just defined, the axioms (TR0)-(TR5) are satisfied.
The axioms (TR0) and (TR2) hold by Definition 6.6.
(TR1) From the mapping cone construction there is a standard triangle
idX
X −→ X −→ M(idX ) −→ X[1].
20 Thorsten Holm and Peter Jørgensen
For constructing an isomorphism between the latter two triangles we take the
identity maps for the first, second and fourth entries. Moreover, we define
morphisms
and conversely
(TR4) Again it suffices to prove the axiom for standard triangles. By assump-
tion we have a diagram
u- α(u)
- M(u) β(u)
- X[1]
X Y
f g f [1]
? u ? α(u ) ?
X - Y - M(u ) β(u-) X [1]
where the left square commutes in K(A), i.e. there exist homotopy maps
Y
sn : Xn → Yn+1 such that gn un − un fn = dn+1 sn + sn−1 dnX for all n ∈ Z. For
completing the diagram to a morphism of triangles we define h = (hn )n∈Z :
M(u) → M(u ) by setting
fn−1 0
hn = : M(u)n = Xn−1 ⊕ Yn → M(u )n = Xn−1
⊕ Yn .
sn−1 gn
This is indeed a morphism of complexes because of the homotopy property of
s given above. Moreover, the completed diagram commutes since by definition
we have that h ◦ α(u) = α(u ) ◦ g and β(u ) ◦ h = f [1] ◦ β(u); note that these
are proper equalities, not only up to homotopy.
(TR5) Again it suffices to prove the octahedral axiom for standard trian-
gles. From the assumptions we already have the following part of the relevant
diagram
X
u -Y α(u)-
M(u)
β(u)-
X[1]
v
?
X
vu - Z α(vu)- M(vu) β(vu)
- X[1]
u u[1]
? ?
Y
v -Z α(v)-
M(v)
β(v)-
Y [1]
We now define the missing morphisms as follows. Let f = (fn ) : M(u) →
idXn−1 0
M(vu) be given in degree n by fn = and set g = (gn ) :
0 vn
un−1 0
M(vu) → M(v) to be given by gn = . Finally define h :
0 idZn
22 Thorsten Holm and Peter Jørgensen
M(v) → M(u)[1] as the composition α(u)[1] ◦ β(v), i.e. it is given by the
0 0
matrix . Then it is easy to check from the definitions that all
idYn−1 0
squares in the completed diagram commute (not only up to homotopy).
For proving (TR5) it now remains to show that the bottom line
f g h
M(u) −→ M(vu) −→ M(v) −→ M(u)[1]
is a distinguished triangle in K(A). To this end we construct an isomorphism
to the standard triangle
f α(f ) β(f )
M(u) −→ M(vu) −→ M(f ) −→ M(u)[1].
Note that only the third entries in the triangles are different. So it suffices
to find morphisms σ = (σn ) : M(v) → M(f ) and τ = (τn ) : M(f ) → M(v)
leading to commutative diagrams (in K(A)!), i.e. we need that β(f ) ◦ σ = h,
h ◦ τ = β(f ), σ ◦ g = α(f ) and τ ◦ α(f ) = g, up to homotopy. Moreover, we
have to show that they are isomorphisms in the homotopy category. We set
⎛ ⎞
0 0
⎜ idYn−1 0 ⎟
σn := ⎜ ⎟ and τn := 0 idYn−1 un−1 0
.
⎝ 0 0 ⎠ 0 0 0 idZn
0 idZn
First, let us check that σ and τ give commutative diagrams. Directly from the
we get that τ ◦ α(f ) = g; in fact both are given in degree n by
definitions
un−1 0
the map : Xn−1 ⊕ Zn → Yn−1 ⊕ Zn . Also by definition we see
0 idZn
0 0
that β(f ) ◦ σ = h, both given by : Yn−1 ⊕ Zn → Xn−2 ⊕ Yn−1 .
idYn−1 0
The remaining commutativities will now only hold up to homotopy. Note that
α(f ) − σ ◦ g : M(vu) → M(f ) is given in degree n by
⎛ ⎞
0 0
⎜ −un−1 0 ⎟
⎜ ⎟ : Xn−1 ⊕ Zn → Xn−2 ⊕ Yn−1 ⊕ Xn−1 ⊕ Zn .
⎝ idX 0⎠
n−1
0 0
We claim that α(f ) − σ ◦ g is homotopic to zero, i.e. α(f ) = σ ◦ g in K(A).
In fact, a homotopy map s = (sn ) where sn : M(vu)n → M(f )n+1 is given by
⎛ ⎞
idXn−1 0
⎜ 0 0⎟
⎜ ⎟ : Xn−1 ⊕ Zn → Xn−1 ⊕ Yn ⊕ Xn ⊕ Zn+1 .
⎝ 0 0⎠
0 0
Triangulated categories 23
For verifying the details recall that the differential of the mapping cone M(f )
is given by
⎛ X ⎞
dn−2 0 0 0
⎜ −un−2 −dn−1 Y
0 0 ⎟
dnM(f ) = ⎜
⎝ idX
⎟.
n−2
0 −d X
0 ⎠
n−1
0 vn−1 (vu)n−1 dnZ
Finally, consider β(f ) − h ◦ τ : M(f ) → M(u)[1] which in degree n is given
by
idXn−2 0 0 0
: Xn−2 ⊕ Yn−1 ⊕ Xn−1 ⊕ Zn → Xn−2 ⊕ Yn−1 .
0 0 −un−1 0
This can be seen to be homotopic to zero by using the homotopy map s = (sn )
where
0 0 idXn−1 0
sn = : Xn−2 ⊕ Yn−1 ⊕ Xn−1 ⊕ Zn → Xn−1 ⊕ Yn .
0 0 0 0
For the straightforward verification again use the differential of M(f ) as given
above.
For completing the proof it now remains to show that σ and τ are iso-
morphisms in the homotopy category. We have τ ◦ σ = idM(v) by definition.
Conversely, the composition σ ◦ τ is in degree n given by
⎛ ⎞
0 0 0 0
⎜ 0 idYn−1 un−1 0 ⎟
⎜ ⎟
⎝0 0 0 0 ⎠
0 0 0 idZn
If we then define homotopy maps sn : M(f )n → M(f )n+1 by setting
⎛ ⎞
0 0 − idXn−1 0
⎜0 0 0 0⎟
sn := ⎜
⎝0 0
⎟
0 0⎠
0 0 0 0
M(f ) M(f )
then we have σ ◦ τ − idM(f ) = dn+1 ◦ sn + sn−1 ◦ dn which is easily
checked using the differential of M(f ) as given above.
Thus σ ◦ τ = idM(f ) in the homotopy category K(A) and we have proved
the octahedral axiom for K(A).
7. Derived categories
A very important class of triangulated categories is formed by derived cat-
egories. They occur frequently in many different areas of mathematics and
have found numerous applications. In this section we shall provide the relevant
constructions leading from the homotopy category to the derived category.
Definition 7.1.
(i) Let X = (Xn , dnX ) be a complex in C(A). The n-th homology of the com-
plex X is defined as the following object from A,
(where the kernel and the image are the usual set-theoretic kernel and
image, respectively).
(ii) The complex X is called exact if Hn (X) = 0 for all n ∈ Z.
(iii) Let f : X → Y be a morphism of complexes. We define an induced map
on the level of homology by setting
Hn (f ) : Hn (X) → Hn (Y ), x + im dn+1
X
→ fn (x) + im dn+1
Y
.
Remark 7.2. Note that the induced map on homology is well-defined; in fact
let x = dn+1
X
(x ) ∈ im dn+1
X
; then
fn (x ) = fn (dn+1
X
(x )) = dn+1
Y
(fn+1 (x )) ∈ im dn+1
Y
.
Hn (f )(x + im dn+1
X
) = fn (x) + im dn+1
Y
= (dn+1
Y
sn + sn−1 dnX + gn )(x) + im dn+1
Y
= gn (x) + im dn+1
Y
= Hn (g)(x + im dn+1
X
).
Hence H (f ) = H (g).
26 Thorsten Holm and Peter Jørgensen
Example 7.5.
(i) (Projective resolutions) As we are restricting in this section to categories R-
Mod of modules over a ring any object X in A has a projective resolution,
i.e. a sequence
. . . → P2 → P1 → P0 → 0
. . . → P2 → P1 → P0 → X → 0.
... → 0 → 0 → X → 0
where X is supposed to be in degree 0. Then
is a quasi-isomorphism.
In fact, in non-zero degrees both complexes have zero homology, and in
degree 0 we have isomorphisms H0 (P ) ∼ =X∼ = H0 (X) induced by
.
(ii) (Injective resolutions) Dually, every object has an injective resolution, i.e.
there is a sequence
0 → I0 → I−1 → I−2 → . . .
where all Ij are injective objects, and a morphism ι : X → I0 such that the
following augmented sequence is exact
ι
0 → X −→ I0 → I−1 → I−2 → . . .
showing well-definedness. The other parts also follow easily from the
definition.
Remark 7.12. The class of quasi-isomorphisms does not in general satisfy the
conditions of the preceding lemma already in C(A); it is crucial first to pass to
the homotopy category. In fact, in the proof below we shall make heavy use of
the triangulated structure of the homotopy category (which has been proven in
Theorem 6.7).
t- uf- w-
W X U W [1]
g f id g[1]
? ? ? ?
q- u- v-
Z Y U Z[1]
Since q is a quasi-isomorphism by assumption, the long exact homology
sequence applied to the bottom row yields that Hn (U ) = 0 for all n ∈ Z. And
then the long exact homology sequence for the top row implies that t must be
a quasi-isomorphism, as desired (cf. Remark 7.8).
The symmetrical second claim in (Q3) is shown similarly.
Finally, let us prove (Q4). We will prove the direction (i)⇒(ii), the con-
verse is proved similarly. For simplicity, set h := f − g, thus q ◦ h = 0
by assumption. By (TR2) and (TR3) there exists a distinguished triangle
u q w
Z −→ Y −→ Y −→ Z[1]. Since q is a quasi-isomorphism, Z[1] and hence Z
is exact (cf. Remark 7.8). By (TR1), (TR3) and (TR4) there exists a morphism
v making the following diagram commutative
id- - 0 - X[1]
X X
v h v[1]
? ? ? ?
u- q - w-
Z Y Y Z[1]
Now again by (TR2) and (TR3) v can be embedded in a distinguished triangle
t v
X −→ X −→ Z −→ X [1]. Here t is a quasi-isomorphism since Z is exact
(cf. Remark 7.8). Moreover, v ◦ t = 0 since the composition of any consecutive
maps in a distinguished triangle vanishes. It follows that h ◦ t = u ◦ v ◦ t = 0,
i.e. f ◦ t = g ◦ t, as desired.
X
q @f
R
@
X Y
We shall use this description frequently in the sequel and hence want to make
this more precise. Again, we follow the approach in the book by Gelfand and
Manin [4, section III.2]. We shall define a category
D(A) where morphisms are
represented by such roofs, and then show that this category is indeed equivalent
to the derived category D(A) introduced in Theorem 7.10.
For computing with these roofs we need to introduce a suitable notion of
equivalence for roofs. Two roofs (q, f ) and (t, g) are called equivalent if there
exists another roof (r, h) making the following diagram commutative.
X
r @h
R
@
X X
q P P g
PPP @
)t f Pq
P R
@
X Y
W
t
@g
R
@
X Y
q @f t @g
R
@ R
@
X Y Z
The category
D(A) is defined as having the same objects as D(A) (and hence
as the homotopy category K(A)), namely complexes over A.
The morphisms in D(A) are defined to be the equivalence classes of roofs,
with the above composition.
Proposition 7.13. Let A be an abelian category. Then the category D(A)
satisfies the universal property of the derived category D(A) given in Theorem
7.10. In particular, the categories D(A) and D(A) are equivalent.
Proof. We first define a functor L : K(A) → D(A) as the identity on objects
and on morphisms by sending f to the roof (id, f ). Clearly, L maps the identity
to the identity. As for composition of morphisms, a composition g ◦ f is on the
one hand sent by L to the roof (id, g ◦ f ); on the other hand the composition
◦ L(f
L(g) ) is given by the roof obtained from the commutative diagram
X
id @f
R
@
X Y
id @f id @g
R
@ R
@
X Y Z
◦ f ) = L(g)
Thus, L(g ◦ L(f ) and L is indeed a functor.
Now it remains to prove that the category D(A), together with the functor
L, satisfies the properties (L1) and (L2) from Theorem 7.10.
For (L1), any quasi-isomorphism q in K(A) is mapped to the roof (id, q). It
is immediate from the above composition of morphisms that (q, id) ◦ (id, q) =
(id, id), and that (id, q) ◦ (q, id) = (q, q); but the latter roof is equivalent
to (id, id). Thus, L maps quasi-isomorphisms to isomorphisms and (L1) is
satisfied.
For proving (L2), let F : K(A) → D (D any category) be a functor which
maps quasi-isomorphisms to isomorphisms. We have to show that there is a
unique functor F : D(A) → D such that F ◦ L = F.
We first deal with uniqueness. On objects X, the only choice is F (X) =
F (X) since L is the identity on objects. Now consider a morphism in D(A),
represented by a roof (q, f ). In D(A) we have that
= (q, f ) ◦ (id, q) = (id, f ) = L(f
(q, f ) ◦ L(q) ).
has to be a functor with F
Since F ◦ L
= F we can deduce that
(L(f
F (f ) = F )) = F ((q, f ) ◦ L(q))
((q, f )) ◦ F
=F (L(q))
=F((q, f )) ◦ F (q).
34 Thorsten Holm and Peter Jørgensen
on
By assumption, F (q) is an isomorphism, so the only possibility to define F
morphisms is to set
((q, f )) = F (f ) ◦ F (q)−1 .
F
X
r @h
R
@
X X
q P P g
PPP @
)t f Pq
P R
@
X Y
implies that
((t, g) ◦ (q, f )) = F
F ((q ◦ t , g ◦ g )) = F (g ◦ g ) ◦ F (q ◦ t )−1
= F (g) ◦ F (g ) ◦ F (t )−1 ◦ F (q)−1
= F (g) ◦ F (t)−1 ◦ F (f ) ◦ F (q)−1
=F((t, g)) ◦ F
((q, f )).
Thus F is indeed a functor and this completes the proof of the universal property
(L2).
Remark 7.14. In the sequel we shall denote the derived category exclusively
by D(A) even if we usually use the more convenient equivalent version
D(A)
just described.
Proof. Following Definition 1.1 we have to show the properties (A1), (A2) and
(A3).
(A1) We first describe addition of morphisms. Let two morphisms F, G
from X to Y be represented by roofs (q, f ) and (q , f ). By the Ore condition
(Q3) there exists an object W , a morphism g and a quasi-isomorphism t making
the following diagram commutative
g
W⏐ −→ X
⏐
t q
q
X −→ X
object X the morphism sets HomD(A) (X, 0) and HomD(A) (0, X) contain only
the morphism represented by the roof (idX , 0) and (0, idX ), respectively. In
fact, any morphism from X to the zero complex is represented by a roof
(q, 0) where q : Z → X is a quasi-isomorphism. But it easily follows from
the definition that the roof (q, 0) is equivalent to (idX , 0), thus HomD(A) (X, 0)
contains precisely one element. The assertion for HomD(A) (0, X) is shown
similarly.
(A3) For the coproduct of two objects X and Y in D(A) one uses the
image of the coproduct X ⊕ Y in K(A) under the localisation functor L
(which is the identity on objects and maps a morphism f in K(A) to the roof
(id, f ) in D(A)). The corresponding maps L(ιX ) : X → X ⊕ Y and L(ιY ) :
X → X ⊕ Y are given by the roofs (id, ιX ) and (id, ιY ), respectively, where
ιX and ιY are the embeddings (or more precisely, their equivalence classes in
K(A)).
We have to show that the universal property (A3) is satisfied. So let f˜X and
˜
fY be arbitrary morphisms in D(A) from X and Y to some object Z. They are
represented by roofs of the form
X̃ Ỹ
q @fX and t @fY
R
@ R
@
X Z Y Z
X̃
q @ ιX̃
@
R
@
X X̃ ⊕ Ỹ
id @
ιX q ⊕ t @ (fX , fY )
@
R
@ @ R
@
X X⊕Y Z
Definition 7.16. The translation functor on D(A) is defined as the shift [1] on
objects, and for a morphism F in D(A) represented by a roof (q, f ) we set
F [1] to be the equivalence class of the roof (q[1], f [1]).
A triangle in D(A) is a distinguished triangle if it is isomorphic (in D(A)!)
to the image of a distinguished triangle from K(A) under the localisation
functor L.
With the above definitions one can then show the main structural property
of derived categories. Unfortunately, the proof of the axioms for a triangulated
category will become very technical so that we shall refrain from providing a
proof in this introductory chapter.
Theorem 7.18. Let A be an abelian category. Then the derived category D(A)
is triangulated.
38 Thorsten Holm and Peter Jørgensen
Proof. The first two implications are obvious. For the third, f : X → Y as
a morphism in D(A) is represented by the roof (idX , f ). For f being 0 in
D(A) means being equivalent to the roof (idX , 0X,Y ), i.e. in K(A) there is a
commutative diagram
Z
r @h
R
@
X X
id PP 0
PPP @
id
) f Pq@
PR
X Y
Remark 7.20.
(1) Note that a morphism f in K(A) becomes zero in D(A) if and only if there
exists a quasi-isomorphism r such that f ◦ r is homotopic to zero.
(2) All implications given in Proposition 7.19 are strict. Let us give examples
for each case. We consider the category A = Ab.
For the first implication, consider the following morphism of complexes
... → 0 → 0 −→ Z⏐ → 0 → ...
id
id
... → 0 → Z −→ Z → 0 → ...
For the second implication, consider the identity map on the (exact)
complex
·2 π
0 → Z −→ Z −→ Z/2Z → 0.
This morphism is not homotopic to zero (i.e. nonzero in K(A)) because
HomZ (Z/2Z, Z) = 0 and hence the identity on Z/2Z can not factor through
a homotopy. However, it is zero in D(A) because we can find a quasi-
isomorphism r such that f ◦ r is homotopic to zero. In fact, let r be the
morphism of complexes
id
0 → 0 → Z⏐ −→ ⏐Z → 0
id π
·2 π
0 → Z −→ Z −→ Z/2Z → 0
Since both complexes are exact, r is a quasi-isomorphism. Moreover,
f ◦ r is homotopic to zero (a homotopy map is given by 0 and id in
the two relevant degrees). This implies that f is zero when considered as
a morphism in D(A).
For the third implication we consider the morphism f : X → Y of
complexes given as follows
·2
0 → Z⏐ −→ ⏐Z → 0
id π
·2
0 → Z −→ Z/3Z → 0
The homology of X is given by H0 (X) = Z/2Z and H1 (X) = 0, whereas
H0 (Y ) = 0 and H1 (Y ) = 3Z (and all other being zero). In particular,
Hn (f ) = 0 for all n ∈ Z.
However, we claim that f is nonzero in the derived category. Suppose
for a contradiction that f = 0 in D(A), i.e. there exist a complex R =
(Rn , dnR ) and a quasi-isomorphism r : R → X such that f ◦ r : R → Y is
homotopic to zero. Since r is a quasi-isomorphism, we have that Hn (R) ∼ =
Hn (X) = 0 for n
= 0 and H0 (R) = H0 (X) ∼ = Z/2Z. Choose a generator
of H0 (R), i.e. z0 ∈ ker d0R \ im d1R . Since r is a quasi-isomorphism, r0 (z0 )
must not be in the image of d1X , i.e. r0 (z0 )
∈ 2Z. On the other hand, f ◦ r
is homotopic to zero, thus there exist homotopy maps s0 : R0 → Z and
s−1 : R−1 → Z/3Z such that (f ◦ r)0 = π ◦ r0 = 2s0 + s−1 ◦ d0R . Applied
to the generator z0 (which is in the kernel of d0R ) this yields
(π ◦ r0 )(z0 ) = (2s0 + s−1 ◦ d0R )(z0 ) = 2s0 (z0 ) ∈ 2(Z/3Z).
But then also r0 (z0 ) ∈ 2Z, a contradiction to the earlier conclusion.
Hence there is no such quasi-isomorphism r, i.e. f
= 0 in D(A).
40 Thorsten Holm and Peter Jørgensen
0 idXn−1 0
π : Cyl(f ) → M(f ) given in degree n by πn = .
0 0 idYn
We leave the straightforward verification to the reader that these maps indeed
commute with the differentials. Clearly, the resulting sequence
ι π
0 → X −→ Cyl(f ) −→ M(f ) → 0
(i) The following diagram with exact rows is commutative in the category
C(A) of complexes.
α(f ) β(f )
0 −→ ⏐Y −→ ⏐ ) −→ X[1]
M(f −→ 0
⏐ ⏐
⏐σ ⏐id
ι π
0 −→ X⏐ −→ ⏐ ) −→
Cyl(f M(f ) −→ 0
⏐ ⏐
⏐id ⏐τ
f
X −→ Y
Proof. (i) It is immediately checked from the definitions (for α(f ) see Remark
6.4) that σ and τ are indeed morphisms of complexes and that all squares in
the diagram commute (in C(A), not only up to homotopy).
(ii) By definition we have τ ◦ σ = idY . On the other hand, σ ◦ τ is homotopic
to the identity via the homotopy map s = (sn ) with
⎛ ⎞
0 0 0
sn = ⎝ idXn 0 0⎠.
0 0 0
f g
Corollary 7.22. To any short exact sequence 0 → X −→ Y −→ Z → 0 in
C(A) there exists a corresponding distinguished triangle in D(A) of the form
f g
X −→ Y −→ Z −→ X[1].
42 Thorsten Holm and Peter Jørgensen
Proof. Using the notations of the previous lemma we consider the following
diagram with exact rows
ι π
0 →X
⏐ −→ Cyl(f
⏐ ) −→ M(f
⏐ ) → 0
⏐ ⏐ ⏐
⏐id ⏐τ ⏐γ
f g
0 →X −→ Y −→ Z → 0
where γ = (γn ) is defined by setting γn (x, y) := gn (y); this is easily checked
to be a morphism of complexes (using that g is a morphism of complexes
and that g ◦ f = 0 since they are consecutive maps in a short exact sequence).
Also one immediately deduces from the definitions that the above diagram
is commutative (where the left hand square already appeared in the previous
lemma).
Since id and τ (by the previous lemma) are both quasi-isomorphisms it
follows from the long exact homology sequences and the usual 5-lemma that
also γ must be a quasi-isomorphism, hence an isomorphism in the derived
category (but not necessarily in K(A), see the following remark). So we have
a morphism of triangles in D(A)
f α(f ) β(f )
X
⏐ −→ ⏐
Y −→ M(f
⏐ ) −→ X[1]
⏐
⏐ ⏐ ⏐ ⏐
⏐id ⏐id ⏐γ ⏐id
f g β(f )◦γ −1
X −→ Y −→ Z −→ X[1]
where the inverse γ −1 exists in the derived category. For the commutativity of
the second square note that by the previous lemma and the above definition of
γ we have
γ ◦ α(f ) = γ ◦ π ◦ σ = g ◦ τ ◦ σ = g ◦ id = g.
is isomorphic to the image of a standard triangle from K(A) under the locali-
sation functor, hence a distinguished triangle in D(A).
Example 8.2.
(i) Every abelian category is an exact category; in fact, take for E the class of
all short exact sequences in A.
(ii) Let A = Ab be the category of abelian groups and B := tf-Ab the full
subcategory of torsionfree abelian groups. Then B is closed under exten-
α β
sions; in fact, let 0 → X → Y → Z → 0 be a short exact sequence with
X and Z torsionfree. Suppose ny = 0 for some y ∈ Y and n ∈ Z. Then
nβ(y) = β(ny) = 0 and hence β(y) = 0 since Z is torsionfree. By exact-
ness of the sequence it follows that y is in the image of α, say y = α(x).
But then α(nx) = nα(x) = ny = 0 which implies nx = 0 since α is a
monomorphism. From the torsionfreeness of X we deduce that x = 0 and
thus also y = 0, i.e. Y is also torsionfree. So the pair (tf-Ab, E) is an exact
category.
However, note that tf-Ab is not an abelian category (e.g. the morphism
·2
Z −→ Z does not have a cokernel in tf-Ab).
(iii) Similarly to the preceding example, consider the category t-Ab with
objects the abelian groups containing torsion elements and the trivial
group. This is a full subcategory of Ab which is closed under extensions;
in fact, if 0 → X → Y → Z → 0 is a short exact sequence of abelian
groups and X and Z contain torsion elements, then Y contains the torsion
elements of X if X is nonzero, and otherwise Y has the torsion elements
of Z. So, (t-Ab, E) is also an exact category.
Example 8.4.
(i) A (finite-dimensional) algebra over a field K is called a Frobenius
algebra if there exists a non-degenerate associative bilinear form on .
Equivalently, if there exists a linear form π : → K such that the kernel
of π does not contain any nonzero left ideal of .
Triangulated categories 45
Example 8.6 (Dual numbers). Let K be a field and consider the algebra
= K[X]/(X 2 ). This 2-dimensional algebra is a Frobenius algebra; in fact,
π(a + bX) := b defines a linear form on such that its kernel does not contain
any nonzero left ideal.
The category of finitely generated -modules is then a Frobenius category. It
has only two indecomposable objects, itself and the one-dimensional simple
module K, where is the only injective (and projective) module.
In the corresponding stable module category several module homomor-
phisms vanish, e.g. we have that Hom(, ) = 0, Hom(, K) = 0 and
Hom(K, ) = 0; on the other hand Hom(K, K) remains 1-dimensional since
the isomorphism can not factor through the injective module .
Our main aim is to describe how the stable category of a Frobenius category
(B, E) carries the structure of a triangulated category. To this end we shall briefly
46 Thorsten Holm and Peter Jørgensen
Lemma 8.7. The morphisms (u) are well-defined in the stable category B,
i.e. they are independent of the choice of the lifting morphisms I (u) and of the
representatives of the morphisms u.
Proof. For a morphism u the morphism (u) has been defined above as indi-
cated in the following diagram
ιX- πX-
X I (X) X
u I (u) (u)
? ? ?
ιY - πY-
Y I (Y ) Y
Now let I (u) and I˜(u) be two liftings for u, with corresponding induced mor-
˜
phisms (u) and (u) from X to Y . Then
((u) − (u))π
˜ X = πY I (u) − πY I (u) = πY σ πX .
˜
Our above construction of the objects X and of the morphisms (u) used
a fixed choice of exact triples X → I (X) → X. The following lemma shows
that this construction does not depend on the choice of the exact triples, more
precisely, a different choice leads to naturally isomorphic functors. In particular,
the object X is uniquely defined up to isomorphism in the stable category B.
ιX πX ιX πX
0 → X −→ I (X) −→ X → 0 and 0 → X −→ I (X) −→ X → 0.
Since I (X) and I (X) are injective in B there are morphisms αX : I (X) →
I (X) and αX : I (X) → I (X) making the left hand squares in the following
diagram commutative; moreover since the rows are short exact sequences these
morphisms induce morphisms βX and βX also making the right hand squares
commutative.
0 -X ιX-
I (X)
πX-
X -0
αX βX
? ?
-X ιX- πX - -0
0 I (X) X
αX
βX
? ?
0 -X ιX-
I (X)
πX-
X -0
Note that since B is closed under extensions the cokernel will again be an object
in B (and not only in the ambient abelian category A).
The images in the stable category B of any triangles of the form
u cιY w
X −→ Y −→ M(u) −→ X
Theorem 8.9. Let (B, E) be a Frobenius category. With the above suspension
functor and the collection of distinguished triangles just defined, the stable
category B is a triangulated category.
50 Thorsten Holm and Peter Jørgensen
References
[1] P. Balmer, Triangulated categories with several triangulations. Available
from the author’s web page at
∼
[2] A. Dold, D. Puppe, Homologie nicht-additiver Funktoren. Ann. Inst. Fourier
Grenoble 11 (1961) 201–312.
[3] R. Farnsteiner, Self-injective algebras: I. The Nakayama permutation, II. Compar-
ison with Frobenius algebras, III. Examples and Morita equivalence, IV. Frobe-
nius algebras and coalgebras. Available at
∼
[4] S.I. Gelfand, Y.I.Manin, Methods of Homological Algebra. Second edition.
Springer Monographs in Mathematics. Springer-Verlag, Berlin, 2003.
[5] D. Happel, Triangulated categories in the representation theory of finite-
dimensional algebras. London Mathematical Society Lecture Note Series, 119.
Cambridge University Press, Cambridge, 1988.
[6] A. Hubery, Notes on the octahedral axiom. Available from the author’s web page
at
∼
[7] M. Kashiwara, P. Schapira, Sheaves on manifolds. Grundlehren der Mathematis-
chen Wissenschaften, 292. Springer-Verlag, Berlin, 1990.
[8] M. Kashiwara, P. Schapira, Categories and sheaves. Grundlehren der Mathema-
tischen Wissenschaften 332. Springer-Verlag, 2006.
[9] B. Keller, On differential graded categories. International Congress of Mathemati-
cians. Vol. II, 151–190, Eur. Math. Soc., Zürich, 2006.
[10] H. Krause, Localization theory for triangulated categories. This volume.
[11] F. Muro, S. Schwede, N. Strickland, Triangulated categories without models.
Invent. Math. 170 (2007), 231-241.
[12] A. Neeman, Triangulated categories. Annals of Mathematics Studies, 148. Prince-
ton University Press, Princeton, NJ, 2001.
[13] A. Neeman, New axioms for triangulated categories. J. Algebra 139 (1991), 221–
255.
[14] S. Schwede, Algebraic versus topological triangulated categories. This volume.
[15] S. Schwede, Torsion invariants for triangulated categories. Preprint (2009), avail-
able from the author’s web page at http://www.math.uni-bonn.de/∼schwede.
[16] J.-L. Verdier, Des catégories dérivées des catégories abéliennes. Astérisque No.
239 (1996).
Triangulated categories 51
Thorsten Holm
Leibniz Universität Hannover, Institut für Algebra, Zahlentheorie und Diskrete
Mathematik, Welfengarten 1, D-30167 Hannover, Germany
E-mail address:
URL:
Peter Jørgensen
School of Mathematics and Statistics, Newcastle University, Newcastle upon Tyne
NE1 7RU, United Kingdom
E-mail address:
URL:
Cohomology over complete intersections
via exterior algebras
luchezar l. avramov and
srikanth b. iyengar
To Karin Erdmann on her 60th birthday.
Introduction
This paper concerns homological invariants of modules and complexes over
complete intersection local rings. The goal is to explain a method by which
one can establish in a uniform way results over such rings by deducing them
from results on DG (that is, differential graded) modules over a graded exterior
algebra, which are often easier to prove. A secondary purpose is to demonstrate
the use of numerical invariants of objects in derived categories, called ‘levels’,
introduced in earlier joint work with Buchweitz and Miller [5]; see Section 1.
Levels allow one to track homological and structural information under changes
of rings or DG algebras, such as those involved when passing from complete
intersections to exterior algebras.
We focus on the complexity and the injective complexity of a complex M
over a complete intersection ring R. These numbers measure, on a polynomial
scale, the rate of growth of the minimal free resolution and the minimal injective
resolution of M, respectively. The relevant basic properties are established in
Section 2.
Complexities can be expressed as dimensions of certain algebraic varieties,
attached to M in [2], and earlier proofs of key results relied on that theory. In
52
Cohomology over complete intersections 53
1. Levels
In this section A denotes a DG algebra.
We write D(A) for the derived category of DG A-modules; it is a triangulated
category with shift functor denoted Σ. The underlying graded object of a DG
object X is denoted X . Rings are treated as DG algebras concentrated in
degree zero; over a ring DG modules are simply complexes, and modules are
DG modules concentrated in degree zero; see [5, §3] for more details and
references. Unless specified otherwise, all DG modules have left actions.
A non-empty subcategory C of D(A) is said to be thick if it is an additive
subcategory closed under retracts, and every exact triangle in D(A) with two
vertices in C has its third vertex in C; thick subcategories are triangulated.
Given a DG A-module X, we write thickA (X) for the smallest thick sub-
category containing X. The existence of such a subcategory can be seen by
realizing it as the intersection of all thick subcategories of D(A) that contain X.
Alternatively, the objects of thickA (X) can be built from X in a series of steps,
described below.
1.1. For every X in D(A) and each n ≥ 0 we define a full subcategory thicknA (X)
of thickA (X), called the nth thickening of X in A, as follows. Set thick0A (X) =
{0}; the objects of thick1A (X) are the retracts of finite direct sums of shifts of X.
For each n ≥ 2, the objects of thicknA (X) are retracts of those U ∈ D(A) that
appear in some exact triangle U → U → U → ΣU with U ∈ thickn−1 A (X)
and U ∈ thick1A (X).
Every thickening of X is clearly embedded in the next one; it is also clear
that their union is a thick subcategory of A containing X, which is therefore
54 L. L. Avramov and S. B. Iyengar
and call it the X-level of U in D(A). It measures the number of extensions needed
to build U out of X. Evidently, levelX A (U ) < ∞ is equivalent to U ∈ thickA (X).
We refer the reader to [5, §§2–6] for a systematic study of levels. The
properties used explicitly in this paper are recorded below.
(1) If thick1A (X) = thick1A (Y ), then one has thicknA (X) = thicknA (Y ) for all n,
and hence an equality levelX A (U ) = levelA (U ).
Y
1.3. Let B be a ring and k a simple B-module. For each B-module H , kB H
denotes the smallest integer l ≥ 0 such that H has a filtration by B-submodules
0 = H 0 ⊆ H 1 ⊆ · · · ⊆ H l−1 ⊆ H l = H
B H = inf{l ∈ N | ml H = 0} ;
see [5, 6.1.3]. If lengthB H is finite, then so is B H , and the converse holds
when H is noetherian; see [5, 6.2(4)].
(3) When H0 (A) = k and the k-module H(U ) is finite, one has an inequality
levelkA (U ) ≤ card{n ∈ Z | Hn (U )
= 0} .
Indeed, (1) and (2) are contained in [5, 6.2(3)], while (3) is extracted from
[5, 6.4].
2. Complexities
In this section we introduce a notion of complexity for DG modules over a
suitable class of DG algebras, and establish some of its elementary properties.
We begin with a reminder of the construction of derived functors on the derived
category of DG modules over a DG algebra; see [6, §1] for details.
2.1. Let A be a DG algebra and Ao its opposite DG algebra.
A semifree filtration of a DG A-module F is a filtration
0 = F 0 ⊆ F 1 ⊆ · · · ⊆ F n−1 ⊆ F n ⊆ · · ·
by DG submodules with n0 F n = F and each DG module F n /F n−1 iso-
morphic to a direct sum of suspensions of A; when one exists F is said to be
semifree. When F is semifree its underlying graded A -module F is free, and
the functors HomA (F, −) and (−⊗A F ) preserve quasi-isomorphisms of DG
A-modules.
A semifree resolution of a DG module U is a quasi-isomorphism F → U
of DG modules with F semifree; such a resolution always exists. If U → V is
a quasi-isomorphism of DG A-modules and G → V is a semifree resolution,
then there is a unique up to homotopy morphism F → G of DG A-modules
such that the composed maps F → G → V and F → U → V are homotopic.
These properties imply that after choosing a resolution FU → U for each
U , the assignments (U, V )
→ HomA (FU , V ) and (W, U )
→ W ⊗A FU define
exact functors
RHomA (−, −) : D(A)o × D(A) → D(Z) and
− ⊗LA − : D(A ) × D(A) → D(Z) ,
o
56 L. L. Avramov and S. B. Iyengar
Lemma 2.2. If U is in thickA (X) for some DG A-module X, then one has
A (U ) = 1; the associated cohomology exact sequence shows that the
and levelX
induction hypothesis implies the desired inequality for complexities.
RHomA (k, U ⊗A0 F ) RHomA (k, U ) ⊗LA0 F RHomA (k, U ) ⊗Lk (k⊗A0 F ) .
Since Hi (k ⊗A0 F ) is finite for each i, and zero for |i| 0 but not for all i,
the equality of injective complexities follows. The argument for complexities
is similar.
(B⊗LA −)
/
D(A) o D(B)
ϕ∗
U ϕ∗ (V ) and B ⊗LA U V .
j : D(A) → D(B) ,
j(k) l
ExtA (U, V ) ∼
= ExtB (jU, jV )
Proof. Set rn = rankk (ExtnA (U, V ) ⊗A0 k). By hypothesis ms = 0 holds for
some integer s, so by Nakayama’s Lemma the A0 -module ExtnA (U, V ) is min-
imally generated by rn elements. Thus, one has surjective homomorphisms
3. Composition products
Let A denote a DG algebra, and let U , V , and W be DG A-modules. In this
section we recall the construction of products in cohomology of DG A-modules.
where |a| denotes the degree of a. For each DG B-module X the complexes
U ⊗k X and Homk (X, U ) have canonical structures of DG module over A ⊗k
B, given by
homomorphism
H(HomA (F, k)) ⊗k H(HomB (G, k)) → H(HomA⊗k B (F ⊗k G, k))
The map F ⊗k G → k ⊗k k = k evidently is a semifree resolution over A ⊗k
B, so the preceding homomorphisms defines a Künneth homomorphism
κ : ExtA (k, k) ⊗k ExtB (k, k) → ExtA⊗k B (k, k) .
It is follows from the definition that this is a homomorphism of k-algebras.
Lemma 3.3. If rankk Ai is finite for each i, then κ is an isomorphism.
Proof. Under the hypothesis on A, one can choose an F that has a semifree
filtration where each free graded A -module (F n /F n−1 ) of finite rank; for
instance, take F to be the classical bar-construction, see [12]. In this case the
map ω from Construction 3.2 is bijective, so H(ω) is an isomorphism.
Construction 4.3. Let A be a DG Hopf algebra over k and set S = ExtA (k, k).
Let V be a DG A-module. Choose semifree resolutions F → k and G → V
over A. The assignment ψ
→ ψ ⊗k G defines a morphism
ζV : S → ExtA (V , V ) .
Proof. We verify the second assertion; the first one follows, as ζk = idS .
We may assume U and V are semifree DG A-modules. Let F → k be a
semifree resolution. By Lemma 4.2, the homology of the complex HomA (F ⊗k
U, F ⊗k V ) is ExtA (U, V ). It implies also that any chain map F ⊗k U →
F ⊗k V is homotopy equivalent to one of the form F ⊗k μ. For any chain map
σ : F → F the compositions
σ ⊗U F ⊗μ
F ⊗k U −−→ F ⊗k U −−→ F ⊗k V
F ⊗μ σ ⊗V
F ⊗k U −−→ F ⊗k V −−→ F ⊗k V
If the k-algebra S = ExtA (k, k) is finitely generated, and the k-vector spaces
H(U ) and H(V ) are finite, then ExtA (U, V ) is finite over S, ExtA (U, U ), and
ExtA (V , V ).
Proof. Since S acts on ExtA (U, V ) through ExtA (U, U ), finiteness over
the former implies finiteness over the latter. The same reasoning applies
with ExtA (U, U ) replaced by ExtA (V , V ), so it suffices to prove finiteness
over S.
We claim that the following full subcategory of D(A) is thick:
Indeed, it is clear that C is closed under retracts and shifts; furthermore, every
exact triangle W → W → W → ΣW in D(A) induces an exact sequence
this is the Hilbert-Serre Theorem, see [15, 4.2] or [13, 13.2]. By [15, 5.5],
the order of the pole at t = 1 of the rational function above is equal to the
Krull dimension of C; that is, to the supremum of the lengths of chains of
homogeneous prime ideals in S containing the annihilator of C. The order
of the pole is independent of the choice of S , so one gets a well defined
notion of Krull dimension of C over S; let dimS C denote this number, and
set dim S = dimS S. By decomposing the rational function above into prime
fractions, see [3, §2] for details, one easily obtains
dimS C = cx ( rankk (Cn )) .
Theorem 4.7. If A is a DG Hopf k-algebra with S = ExtA (k, k) finitely gen-
erated as k-algebra, and U a DG A-module with H(U ) finite over k, then one
has
inj cxA U = cxA (U, U ) = cxA U ≤ cxA k = dim S .
If, furthermore, V is a DG A-module with H(V ) finite over k, then one has
dimS (ExtA (U, V )) = cxA (U, V ) ≤ min{cxA (U, U ), cxA (V , V )} .
Proof. The expression for cxA (U, V ) comes from Propositions 4.4, 4.5, and
4.6. By Proposition 4.5, ExtA (U, V ) is a finite module over ExtA (U, U ) and
ExtA (V , V ), whence the upper bounds on cxA (U, V ). For (k, U ) and (U, k)
they yield
inj cxA U ≤ cxA (U, U ) ≥ cxA U ≤ cxA k .
To prove cxA (U, U ) ≤ cxA U we show that the full subcategory
D = {W ∈ D(A) | cxA (U, W ) ≤ cxA U }
of D(A) contains U . One evidently has k ∈ D, and 1.4(1) gives U ∈ thickA (k),
so it suffices to prove that D is thick. Closure under direct summands and shifts
is clear. An exact triangle W → W → W → ΣW in D(A) yields an exact
sequence
ExtnA (U, W ) → ExtnA (U, W ) → ExtnA (U, W )
for every n ∈ Z. They imply inequalities
rankk ExtnA (U, W ) ≤ rankk ExtnA (U, W ) + rankk ExtnA (U, W ) .
Thus, if W and W are in D, then, Lemma 2.6 gives
cxA (U, W ) ≤ max{cxA (U, W ), cxA (U, W )} ≤ cxA U .
A similar argument, now using ExtA (−, U ), yields inj cxA U = cxA (U, U ).
64 L. L. Avramov and S. B. Iyengar
5. Exterior algebras I.
In this section k is a field and a DG algebra with ∂ = 0 and with
an exterior k-algebra on alternating indeterminates ξ1 , . . . , ξc of positive odd
degrees.
Proof. One has ∼ = kξ1 ⊗k B, where kξj denotes the exterior algebra on
the graded vector space kξi and B = kξ2 ⊗k · · · ⊗k kξc . By Lemma 3.3
and induction, it suffices to treat the case = kξ . We use the notation from
Construction 3.1.
Set d = |ξ |, and let F be a DG -module, whose underlying graded -
module has a basis {en }n0 with |en | = (d + 1)n, and with differential defined
by the formulas ∂(e0 ) = 0 and ∂(en ) = ξ en−1 for n ≥ 1. An elementary verifi-
cation shows that the -linear map ε : F → k with ε(e0 ) = 1 and ε(en ) = 0 for
n ≥ 1 is a quasi-isomorphism of DG -modules, and thus a semifree resolution
of k. It yields
[χ (j ) ][χ (i) ] = [χ (j ) ◦ χ
(i) ] = [χ (i+j ) ] .
for all i, j ≥ 0. They show that the isomorphism of graded k-vector spaces
k[χ ] → S, which sends χ i to [χ (i) ] for each i ≥ 0, is a homomorphism of
graded k-algebras.
Remark 5.2. The universal property of exterior algebras guarantees that there
is a unique homomorphism of graded k-algebras : → ⊗k with
(ξi ) = ξi ⊗ 1 + 1 ⊗ ξi for 1 ≤ i ≤ c.
In view of the preceding proposition and remark, Theorems 4.7 and 4.9
yield:
Theorem 5.3. For DG -modules U, V with H(U ) and H(V ) finite over k one
has
k : D(R) → D(K) .
Lemma 6.2. For each homologically finite complex of R-modules M one has:
so the local ring Q/(q) is regular along with Q. The map κ factors through a
homomorphism Q/(q) → R, and the latter is a minimal Cohen presentation.
where the last isomorphism holds because one has j(k) k by Lemma 2.5.
We come to the main result of this section. All its assertions are known: the
first equality is proved in [2, 5.3], the inequality follows from [8, 4.2], and the
second equality from [16, Thm. 6]. The point here is that they are deduced, in
a uniform way, from the corresponding relations for DG modules over exterior
algebras, established in Theorem 5.3, which ultimately are much simpler to
prove.
Theorem 6.7. If (R, m, k) is complete intersection, then for every complex of
R-modules M with H(M) finitely generated the following inequalities hold:
inj cxR M = cxR M ≤ cxR k = codim R .
Proof. The isomorphism of Lemma 6.6 implies cx k = cx jk(k). Now
Lemma 6.5 translates part of Theorem 5.3 into the desired statement.
The equalities in the theorem may fail when R is not complete intersection:
Remark 6.8. Gulliksen [9, 2.3] proved that the condition cxR k < ∞ charac-
terizes local complete intersection rings among all local rings.
Jorgensen and Şega [10, 1.2] construct Gorenstein local k-algebras R with
rankk R finite and modules M with {cxR M, inj cxR M} = {1, ∞}, in one order
or the other.
Remark 6.9. The unused portion of Theorem 5.3 suggests that the relations
cxR (N, M) = cxR (M, N) ≤ cxR (M, M) = cxR M
70 L. L. Avramov and S. B. Iyengar
might hold also over complete intersections. They do, see [4, 5.7], but we know
of no simple way to deduce them from Theorem 5.3.
7. Projective levels
In this section A is a DG algebra. We collect some results on A-levels of
DG modules, which are reminiscent of theorems on projective dimension for
modules.
7.1. For any DG A-module U the following hold.
(1) One has levelA A (U ) ≤ l + 1 if and only if U is isomorphic in D(A) to a
direct summand of some semifree DG module F with a semifree filtration
having F l = F and every F n /F n−1 of finite rank over A ; see 2.1.
(2) When A is non-negative, H0 (A) is a field, and Hi (U ) = 0 for all i 0,
one has levelAA (U ) < ∞ if and only if Tor (k, U ) is finite over k.
A
(3) When A has zero differential and is noetherian, and the graded A-module
H(U ) is finitely generated, the following inequalities hold:
Let Df () be the full subcategory of D() whose objects are the DG modules
U with H(U ) finite over , and Df (S) the corresponding subcategory of D(S).
The next proposition refines the equivalences in (8.1.1); see Remark 8.3.
Proposition 8.2. For each U ∈ Df () and each M ∈ Df (S) one has
Proof. One has Df () = thick (k) by 1.4(1). As the polynomial ring S has
finite global dimension, 7.1(3) gives Df (S) = thickS (S). Thus, the top row of
(8.1.1) gives inverse equivalences between Df () and Df (S). It is easy to verify
that the subcategories in the lower row are thick, so it suffices to compute cx U
and dimS H(M).
The equivalence h gives the first isomorphism in the chain
the second one comes from (8.1.2), the third is clear. In homology, one obtains
Extn (k, U ) ∼
= Hn (h(U ))
72 L. L. Avramov and S. B. Iyengar
for every n ∈ Z, whence the second equality in the following sequence; the
first one comes from Theorem 5.3, the third from the Hilbert-Serre Theorem,
see 4.6:
Remark 8.3. For U ∈ Df () the definition of complexity shows that cx U ≤ 0
is equivalent to the finiteness of the k-vector space Ext (U, k). It is the graded
k-dual of Tor (U, k), so by 7.1(2) its finiteness is equivalent to U ∈ thick ().
On the other hand, M ∈ Df (S) has dimS H(M) ≤ 0 if and only if M is in
thickS (k); see 1.4(1). Thus, for n = 0 diagram (8.2.1) reduces to (8.1.1).
Theorem 8.4. If U is a DG -module with 0 < rankk H(U ) < ∞, then one
has
Proof. The inequality on the left comes from 1.4(3). The one on the right results
from the following chain of (in)equalities
The first one holds because h is an equivalence, see 1.2(2), the second comes
from (8.1.2), and the third from 7.2; the remaining equalities hold by [13,
Exercise 5.1]1 , by Lemma 8.2, and because dim S = c holds.
Example 8.5. For each integer i with 0 ≤ i ≤ c and (i) = /(ξ1 , . . . , ξi ) one
has
Indeed, the first one is clear. For (i) = /(ξi+1 , . . . , ξn ) one has
Extn ((i) , k) ∼
= Extn(i) ⊗k (i) ((i) , k) ∼
= Extn(i) (k, k) ,
D(R)
k / D(K) j
/ D()
≡
the second of which is an equivalence. One then has the following relations
jk(k)
levelkR (M) ≥ level (jk(M))
= levelk (jk(M))
≥ c − cx (jk(M)) + 1
= codim R − cxR M + 1
where the inequalities are given by 1.2(2) and by Theorem 8.4, while the
equalities come from 1.2(1) and Lemmas 6.6, and from Lemma 6.5.
74 L. L. Avramov and S. B. Iyengar
References
[1] C. Allday, V. Puppe, Cohomological methods in transf ormation groups, Cam-
bridge Stud. Adv. Math. 32, Cambridge Univ. Press, Cambridge, 1993.
[2] L. L. Avramov, Modules of finite virtual projective dimension, Invent. Math. 96
(1989), 71–101.
Cohomology over complete intersections 75
Contents
1. Introduction 76
2. An informal introduction to cluster-finite cluster algebras 79
3. Symmetric cluster algebras without coefficients 83
4. Cluster algebras with coefficients 92
5. Categorification via cluster categories: the finite case 99
6. Categorification via cluster categories: the acyclic case 122
7. Categorification via 2-Calabi-Yau categories 126
8. Application: The periodicity conjecture 138
9. Quiver mutation and derived equivalence 143
References 154
1. Introduction
1.1. Context
Cluster algebras were invented by S. Fomin and A. Zelevinsky [50] in the
spring of the year 2000 in a project whose aim it was to develop a combinato-
rial approach to the results obtained by G. Lusztig concerning total positivity in
algebraic groups [103] on the one hand and canonical bases in quantum groups
[102] on the other hand (let us stress that canonical bases were discovered
76
Cluster algebras, representations, triangulated categories 77
We refer to the introductory papers [53] [132] [134] [135] [136] and to the
cluster algebras portal [47] for more information on cluster algebras and their
links with other parts of mathematics.
The link between cluster algebras and quiver representations follows the
spirit of categorification: One tries to interpret cluster algebras as combinatorial
(perhaps K-theoretic) invariants associated with categories of representations.
Thanks to the rich structure of these categories, one can then hope to prove
results on cluster algebras which seem beyond the scope of the purely combina-
torial methods. It turns out that the link becomes especially beautiful if we use
triangulated categories constructed from categories of quiver representations.
1.2. Contents
We start with an informal presentation of Fomin-Zelevinsky’s classification the-
orem and of the cluster algebras (without coefficients) associated with Dynkin
diagrams. Then we successively introduce quiver mutations, the cluster algebra
associated with a quiver, and the cluster algebra with coefficients associated
with an ‘ice quiver’ (a quiver some of whose vertices are frozen). We illustrate
78 Bernhard Keller
Acknowledgments
These notes are based on lectures given at the IRTG-Summerschool 2006
(Schloss Reisensburg, Bavaria) and at the Midrasha Mathematicae 2008
(Hebrew University, Jerusalem). I thank the organizers of these events for
their generous invitations and for providing stimulating working conditions. I
am grateful to Thorsten Holm, Peter Jørgensen and Raphael Rouquier for their
encouragment and for accepting to include these notes in the proceedings of
the ‘Workshop on triangulated categories’ they organized at Leeds in 2006.
It is a pleasure to thank to Carles Casacuberta, André Joyal, Joachim Kock,
Amnon Neeman and Frank Neumann for an invitation to the Centre de Recerca
Matemàtica, Barcelona, where most of this text was written down. I thank
Lingyan Guo and Sefi Ladkani for kindly pointing out misprints and inaccu-
racies. I am indebted to Tom Bridgeland, Bernard Leclerc, David Kazhdan,
Tomoki Nakanishi, Raphaël Rouquier and Michel Van den Bergh for helpful
conversations.
Cluster algebras, representations, triangulated categories 79
= A2 : ◦ ◦.
associated
Now we draw the so-called repetition (or Bratteli diagram) Z
made up of a countable number of
with : We first draw the product Z ×
80 Bernhard Keller
are the 5 cluster variables x1 , x2 , x1 , x2 and x1 and the cluster algebra AA2 is the
Q-subalgebra (not the subfield!) of Q(x1 , x2 ) generated by these 5 variables.
Before going on to a more complicated example, let us record the remarkable
phenomena we have observed:
(1) All denominators of all cluster variables are monomials. In other words,
the cluster variables are Laurent polynomials. This Laurent phenomenon
holds for all cluster variables in all cluster algebras, as shown by Fomin
and Zelevinsky [51].
(2) The computation is periodic and thus only yields finitely many cluster
variables. Of course, this was to be expected by the classification theorem
above. In fact, the procedure generalizes easily from Dynkin diagrams to
arbitrary trees, and then periodicity characterizes Dynkin diagrams among
trees.
(3) Numerology: We have obtained 5 cluster variables. Now we have 5 = 2 + 3
and this decomposition does correspond to a natural partition of the set of
cluster variables into the two initial cluster variables x1 and x2 and the
three non initial ones x1 , x2 and x1 . The latter are in natural bijection with
the positive roots α1 , α1 + α2 and α2 of the root system of type A2 with
simple roots α1 and α2 . To see this, it suffices to look at the denominators
of the three variables: The denominator x1d1 x2d2 corresponds to the root
d1 α1 + d2 α2 . It was proved by Fomin-Zelevinsky [52] that this generalizes
to arbitrary Dynkin diagrams. In particular, the number of cluster variables
in the cluster algebra A always equals the sum of the rank and the number
of positive roots of .
Let us now consider the example A3 : We choose the following linear orientation:
1 /2 / 3.
α2 gOO α +α 2α1D + α2 3α 1 + α2
OOO 1 Z44 2
o o7
OOO 44
o
OOO 4
ooo
OOO44
ooooo
O
oo
−α1 o o o
4O4OOO / α1
o o o
4 O O
ooo
44 OOO
ooo 44 OOO
oo
4 OO'
o o
wo
−α2
Cluster algebras, representations, triangulated categories 83
: 1
(3,1)
/ 2.
The mutation rule is a variation on the one we are already familiar with: In
the recursion formula, each predecessor p of a cluster variable x has to be
raised to the power indicated by the valuation ‘closest’ to p. Thus, we have for
example
3.1. Quivers
Let us recall that a quiver Q is an oriented graph. Thus, it is a quadruple given
by a set Q0 (the set of vertices), a set Q1 (the set of arrows) and two maps
s : Q1 → Q0 and t : Q1 → Q0 which take an arrow to its source respectively
its target. Our quivers are ‘abstract graphs’ but in practice we draw them as in
84 Bernhard Keller
this example:
$ //
Q:
^=== μ
3 α 5 /6
λ ==
==
β
/
1 /2o 4.
ν
γ
In the exchange relation (3.2.1), if there are no arrows from i with target k,
the product is taken over the empty set and equals 1. It is not hard to see that
μk (R, u) is indeed a seed and that μk is an involution: we have μk (μk (R, u)) =
(R, u). Notice that the expression given in (3.2.1) for uk is subtraction-free.
To a quiver R without loops or 2-cycles with vertex set {1, . . . , n} there
corresponds the n × n antisymmetric integer matrix B whose entry bij is the
number of arrows i → j minus the number of arrows j → i in R (notice
that at least one of these numbers is zero since R does not have 2-cycles).
Clearly, this correspondence yields a bijection. Under this bijection, the matrix
B corresponding to the mutation μk (R) has the entries
−bij if i = k or j = k;
bij =
bij + sgn(bik )[bik bkj ]+ else,
where [x]+ = max(x, 0). This is matrix mutation as it was defined by Fomin-
Zelevinsky in their seminal paper [50], cf. also [55].
1 (3.3.1)
E 222
22
22
2
2 o 3
and u = {x1 , x2 , x3 }. If we mutate at k = 1, we obtain the quiver
1
Y333
33
33
3
2 3
and the set of fractions given by u1 = (x2 + x3 )/x1 , u2 = u2 = x2 and u3 =
u3 = x3 . Now, if we mutate again at 1, we obtain the original seed. This is a
general fact: Mutation at k is an involution. If, on the other hand, we mutate
86 Bernhard Keller
1
E X111
11
11
2 3
and the set u given by u1 = u1 = (x2 + x3 )/x1 , u2 = x1 +x 2 +x3
x1 x2
and u3 = u3 =
x3 .
An important special case of quiver mutation is the mutation at a source (a
vertex without incoming arrows) or a sink (a vertex without outgoing arrows).
In this case, the mutation only reverses the arrows incident with the mutating
vertex. It is easy to see that all orientations of a tree are mutation equivalent and
that only sink and source mutations are needed to pass from one orientation to
any other.
Let us consider the following, more complicated quiver glued together from
four 3-cycles:
1 (3.3.2)
F 22
2
o
E 2 : A 3 33
sg ggg 5 kWWWW3
4 6.
1 1 1 1
F 222 E X111 O 222 O 222
2 ]; 3 2 ]; B 3 2 3 2 3
3 WW
o ggg 5 WW+ ggg 3 WWWW+ 3 WW
o ggg 5 WW+ 6 5 kWWWW
4g
g g 5
4 6 4 o 6 4 __________/ 6.
Notice that the last quiver no longer has any oriented cycles and is in fact an
orientation of the Dynkin diagram of type D6 . The sequence of new fractions
appearing in these steps is
x3 x4 +x2 x6 x3 x4 +x1 x5 +x2 x6
u5 = , u3 = ,
x5 x3 x5
x2 x3 x4 +x32 x4 +x1 x2 x5 +x22 x6 +x2 x3 x6 x3 x4 + x4 x5 + x2 x6
u1 = , u6 = .
x1 x3 x5 x5 x6
Cluster algebras, representations, triangulated categories 87
It is remarkable that all the denominators appearing here are monomials and
that all the coefficients in the numerators are positive.
Finally, let us consider the quiver
E12 (3.3.3)
222
o
F 2 33 F 3 22
33 22
o o
4
F 22 E 5 22 F 6 22
22 22 22
7o 8o 9o 10.
One can show [91] that it is impossible to transform it into a quiver without
oriented cycles by a finite sequence of mutations. However, its mutation class
(the set of all quivers obtained from it by iterated mutations) contains many
quivers with just one oriented cycle, for example
10 \\\- 5
6 } 5 jUUU 3
46 5/
!! /
! ~} Q$$
$
1
7 m\\\ 6 x 6 8
qqq ::
x 7
Z55 |x
7
7$
$$ xx<
2 ooo \99
6
10 NN
8 RRR( lll5 4 X1 10 9 xrr &
> EE" 8* 4#
9! 1 8 UUU* || ** ##
!! 9 1 1
3
3 2.
2
In fact, in this example, the mutation class is finite and it can be completely com-
puted using, for example, [87]: It consists of 5739 quivers up to isomorphism.
The above quivers are members of the mutation class containing relatively few
arrows. The initial quiver is the unique member of its mutation class with the
largest number of arrows. Here are some other quivers in the mutation class
with a relatively large number of arrows:
◦
C -- B ◦- n]]]] ◦ > ◦& jVVVV
- --- E 111 ||| &&& ~? ◦''
◦- m\\\ ◦ m[[[ - 1 ◦& iTTTT ~~ ''
-- C ,,, C ◦. ◦ lYYYYY p````` ◦ && ◦ kWW
◦H 7 G & {{= ''' WW? ◦(
, .. 77 { ' ~ (
◦ mZZZ ◦ n]]]] m[[ 7 ◦ jUUUU ~~ ((
.. C ◦- [ ◦ ◦ qbbbb ◦H 6 ◦ kWWWW
. -- C
H 777 666 ◦ Z4
◦ m\\\ ◦ 7 444
qbbbb ◦ qcccc ◦
◦ ◦ aaaa0 ◦
88 Bernhard Keller
Only 84 among the 5739 quivers in the mutation class contain double
arrows (and none contain arrows of multiplicity ≥ 3). Here is a typical
example
6 1 PPPP
mmmm ' ``````0 4
ww I R%
3I 6
J ww %%%
w2 %
8 X1 {www aaaaaa0
11 10 QaQaQa 2
Q(
7 @5
9
The classification of the quivers with a finite mutation class is still open. Many
examples are given in [49] and [38].
The quivers (3.3.1), (3.3.2) and (3.3.3) are part of a family which appears in
the study of the cluster algebra structure on the coordinate algebra of the sub-
group of upper unitriangular matrices in SL(n, C), cf. section 4.6. The quiver
(3.3.3) is associated with the elliptic root system E8(1,1) in the notations of
Saito [118], cf. Remark 19.4 in [64]. The study of coordinate algebras on vari-
eties associated with reductive algebraic groups (in particular, double Bruhat
cells) has provided a major impetus for the development of cluster algebras,
cf. [17].
of the vertices and the associated cluster variables and whose edges correspond
to mutations.
A remarkable theorem due to Gekhtman-Shapiro-Vainshtein states that each
cluster u occurs in a unique seed (R, u), cf. [68].
Notice that the knitting algorithm only produced the cluster variables
whereas this definition yields additional structure: the clusters.
(x1 ← x2 )
kk SSS
kkkkkkk SSS
SSS
kk SSS
kkkk SS
(x1 → xG2 ) (x1 → x2 )
GG ww
GG ww
GG w
GG w
ww
(x1 → x1 )
(x1 ← x2 )
where we have written x1 → x2 for the seed (1 → 2, {x1 , x2 }). Notice that it is
not possible to find a consistent labeling of the edges by 1’s and 2’s. The reason
for this is that the vertices of the exchange graph are not the seeds but the seeds
up to renumbering of vertices and variables. Here the clusters are precisely the
pairs of consecutive variables in the cyclic ordering of x1 , . . . , x1 .
Q: 1 /2 /3
In fact, as it is easy to check, each slice yields a cluster. However, some clusters
do not come from slices, for example the cluster x1 , x3 , x1 associated with the
seed μ2 (Q, x).
a) All cluster variables are Laurent polynomials, i.e. their denominators are
monomials.
b) The number of cluster variables is finite iff Q is mutation equivalent to an
orientation of a simply laced Dynkin diagram . In this case, is unique
and the non initial cluster variables are in bijection with the positive roots
of ; namely, if we denote the simple roots by α1 , . . . , αn , then for each
positive root d α , there is a unique non initial cluster variable whose
i i
denominator is xidi .
c) The knitting algorithm yields all cluster variables iff the quiver Q has two
vertices or is an orientation of a simply laced Dynkin diagram .
The theorem can be extended to the non simply laced case if we work with
valued quivers as in the example of G2 in section 2.2.
It is not hard to check that the knitting algorithm yields exactly the cluster
variables obtained by iterated mutations at sinks and sources. Remarkably, in
the Dynkin case, all cluster variables can be obtained in this way.
The construction of the cluster algebra shows that if the quiver Q is mutation-
equivalent to Q , then we have an isomorphism
∼
AQ → AQ
preserving clusters and cluster variables. Thus, to prove that the condition in b)
is sufficient, it suffices to show that AQ is cluster-finite if the underlying graph
of Q is a Dynkin diagram.
No normal form for mutation-equivalence is known in general and it is
unkown how to decide whether two given quivers are mutation-equivalent.
However, for certain restricted classes, the answer to this problem is known:
Trivially, two quivers with two vertices are mutation-equivalent iff they are
isomorphic. But it is already a non-trivial problem to decide when a quiver
1 ^<
r /2,
<<
s t
3
where r, s and t are non negative integers, is mutation-equivalent to a quiver
without a 3-cycle: As shown in [15], this is the case iff the ‘Markoff inequality’
r 2 + s 2 + t 2 − rst > 4
1o 3o 5o .? . . o n< − 1
@ @ ~~~ xx
~~ xx
~~ xx
~~ xxx
2o 4o 6o ... o n
is cluster-finite of respective cluster-type A3 , D4 , D5 , E6 , E7 and E8 and that
it is not cluster-finite if n > 8.
4.1. Definition
with
Let 1 ≤ n ≤ m be integers. An ice quiver of type (n, m) is a quiver Q
vertex set
{1, . . . , m} = {1, . . . , n} ∪ {n + 1, . . . , m}
such that there are no arrows between any vertices i, j which are strictly greater
than n. The principal part of Q is the full subquiver Q of Q whose vertex set is
{1, . . . , n} (a subquiver is full if, with any two vertices, it contains all the arrows
between them). The vertices n + 1, . . . , m are often called frozen vertices. The
cluster algebra
AQ ⊂ Q(x1 , . . . , xm )
– only mutations with respect to vertices in the principal part are allowed and
no arrows are drawn between the vertices > n,
Cluster algebras, representations, triangulated categories 93
– in a cluster
u = {u1 , . . . , un , xn+1 , . . . , xm }
only u1 , . . . , un are called cluster variables; the elements xn+1 , . . . , xm are
called coefficients; to make things clear, the set u is often called an extended
cluster;
– the cluster type of Q is that of Q if it is defined.
where we have framed the frozen vertices. Indeed, here the principal part Q
only consists of the vertex 1 and we can only perform one mutation, whose
associated exchange relation reads
x1 x1 = 1 + x2 x3 or x1 x1 − x2 x3 = 1.
We obtain an isomorphism as required by sending x1 to a, x1 to d, x2 to b and
x3 to c. We describe this situation by saying that the quiver
a>
>>>
>>
>
b c
94 Bernhard Keller
Each plane P in Cn+3 gives rise to a straight line in this cone, namely the one
generated by the 2 × 2-minors xij of any (n + 3) × 2-matrix whose columns
generate P . Notice that the monomials in the Plücker relation are naturally
associated with the sides and the diagonals of the square
i << j
O << O
O< O
<< O
O
<
l k
The relation expresses the product of the variables associated with the diagonals
as the sum of the monomials associated with the two pairs of opposite sides.
Now the idea is that the Plücker relations are exactly the exchange relations
for a suitable structure of cluster algebra with coefficients on the coordinate
ring. To formulate this more precisely, let us consider a regular (n + 3)-gon in
the plane with vertices numbered 1, . . . , n + 2, and consider the variable xij as
associated with the segment [ij ] joining the vertices i and j .
Proposition 4.1 ([52, Example 12.6]). The algebra A has a structure of cluster
algebra with coefficients such that
– the coefficients are the variables xij associated with the sides of the (n + 3)-
gon;
– the cluster variables are the variables xij associated with the diagonals of
the (n + 3)-gon;
– the clusters are the n-tuples of variables whose associated diagonals form a
triangulation of the (n + 3)-gon.
Moreover, the exchange relations are exactly the Plücker relations and the
cluster type is An .
Cluster algebras, representations, triangulated categories 95
o 0 OO
ooo 44 OO
ooo 44 OO
05 01
W//
5 44 1
44 45 / 04 gOOO 02 / 12
44 wooo O
4
03
? ???
4 NNN
NNN oooooo
2
o 34 23
3
Theorem 4.2 ([69]). The algebra A has the structure of a cluster algebra with
coefficients whose initial seed is given by
96 Bernhard Keller
The following table indicates when these algebras are cluster-finite and what
their cluster-type is:
k\n 2 3 4 5 6 7
2 A1 A2 A3 A4 A5 A6
3 A2 D4 E6 E8
4 A3 E6
5 A4 E8
where the xij are the coordinate functions on SL(n, C). This bracket makes
G = SL(n, C) into a Poisson-Lie group and P \ G into a Poisson G-variety for
each subgroup P of G containing the subgroup B of lower triangular matrices.
In particular, the big cell of the Grassmannian considered above inherits a
Poisson bracket.
Theorem 4.3 ([69]). This bracket is compatible with the cluster algebra struc-
ture in the sense that each extended cluster is a log-canonical coordinate system,
Cluster algebras, representations, triangulated categories 97
i.e. we have
ω(ui , uj ) = ωij(u) ui uj
for certain (integer) constants ωij(u) depending on the extended cluster u. More-
over, the coefficients are central for ω.
be an ice quiver.
This theorem admits the following generalization: Let Q
to be obtained by glueing the complex tori
Define the cluster variety X (Q)
indexed by the clusters u
using the exchange relations as glueing maps, where m is the number of vertices
of Q.
Theorem 4.4 ([69]). Suppose that the principal part Q of Q is connected and
that the matrix B associated with Q is of maximal rank. Then the vector space
compatible with the cluster algebra structure is
of Poisson structures on X (Q)
of dimension
m−n
1+ .
2
is an open subset of the
Notice that in general, the cluster variety X (Q)
spectrum of the (complexified) cluster algebra. For example, for the cluster
algebra associated with SL(2, C) which we have considered above, the cluster
variety is the union of the elements
a b
c d
of SL(2, C) such that we have abc
= 0 or bcd
= 0. The cluster variety is
always regular, but the spectrum of the cluster algebra may be singular.
For example, the spectrum of the cluster algebra associated with the ice
quiver
x@
~~~ @@@ 2
~~ @@
2
~
~ @
u v
Theorem 4.5 ([17]). The coordinate algebra C[N ] has an upper cluster alge-
bra structure whose initial seed is given by
.. / ...
.O
~
fn,2
We refer to [17] for the notion of ‘upper’ cluster algebra structure. It is not
hard to check that this structure is of cluster type A3 for n = 3, D6 for n = 4
and cluster-infinite for n ≥ 5. For n = 5, this cluster algebra is related to the
elliptic root system E8(1,1) in the notations of Saito [118], cf. [64].
A theorem of Fekete [42] generalized in [16] claims that a square matrix
of order n + 1 is totally positive (i.e. all its minors are > 0) if and only if the
following (n + 1)2 minors of g are positive: all minors occupying several initial
rows and several consecutive columns, and all minors occupying several initial
columns and several consecutive rows. It follows that an element g of N is
totally positive if fij (g) > 0 for the fij belonging to the initial seed above.
The same holds for the u1 , . . . , um in place of these fij for any cluster u of
Cluster algebras, representations, triangulated categories 99
this cluster algebra because each exchange relation expresses the new variable
subtraction-free in the old variables.
Geiss-Leclerc-Schröer have shown [65] that each monomial in the variables
of an arbitrary cluster belongs to Lusztig’s dual semicanonical basis of C[N ]
[104]. They also show that the dual semicanonical basis of C[N ] is different
from the dual canonical basis of Lusztig and Kashiwara except in types A2 , A3
and A4 [64].
α ttt
9 2 KKK β
t KKK
tt %/
1 γ 3.
rr9
V2 LL V
r
Vα
r LLLβ
rr L%
V1 / V3
Vγ
Vi
Vα
/ Vj
fi fj
Wi / Wj
Wα
1 /2,
and
V : V1
Vα
/ V2
VO 1
Vα
/ V2
O
kn / kp ,
A
kO n / kp
A O
Q P −1
kn / kp
Ir ⊕0
(V ⊕ W )i = Vi ⊕ Wi and (V ⊕ W )α = Vα ⊕ Wα ,
for all vertices i and all arrows α of Q. For example, the above representation
in normal form is isomorphic to the direct sum
(k
1 / k )r ⊕ ( k / 0 )n−r ⊕ ( 0 / k )p−r .
ker(f )i = ker(fi : Vi → Wi )
endowed with the maps induced by the Vα and similarly for the cokernel. A
subrepresentation V of a representation V is given by a family of subspaces
Vi ⊂ Vi , i ∈ Q0 , such that the image of Vi under Vα is contained in Vj for
each arrow α : i → j of Q. A sequence
0 /U /V /W /0
0 / Ui / Vi / Wi /0
k / 0 and 0 /k
V = (k
1 / k)
0 → (0 / k) → (k 1 / k) → (k / 0) → 0
As we have seen above, for quivers without oriented cycles, the classifica-
tion of the simple representations is trivial. On the other hand, the problem of
classifying the indecomposable representations is non trivial. Let us examine
this problem in a few examples: For the quiver 1 → 2, we have checked the
existence in part b) directly. The uniqueness in b) then implies that each inde-
composable representation is isomorphic to exactly one of the representations
S1 , S2 and
k
1 / k.
Similarly, using elementary linear algebra it is not hard to check that each
indecomposable representation of the quiver
An : 1 /2 / ... /n
1d α
Question 5.2. For which quivers are there only finitely many isomorphism
classes of indecomposable representations?
qQ : ZQ0 → Z
by
qQ (v) = vi2 − vs(α) vt(α) .
i∈Q0 α∈Q1
Notice that the Tits form does not depend on the orientation of the arrows of Q
but only on its underlying graph. We say that the quiver Q is representation-
finite if, up to isomorphism, it has only finitely many indecomposable repre-
sentations. We say that a vector v ∈ ZQ0 is a root of qQ if qQ (v) = 1 and that
it is positive if its components are ≥ 0.
Theorem 5.3 (Gabriel [58]). Let Q be a connected quiver and assume that k
is algebraically closed. The following are equivalent:
104 Bernhard Keller
(i) Q is representation-finite;
(ii) qQ is positive definite;
(iii) The underlying graph of Q is a simply laced Dynkin diagram .
Moreover, in this case, the map taking a representation to its dimension vec-
tor yields a bijection from the set of isomorphism classes of indecomposable
representations to the set of positive roots of the Tits form qQ .
It is not hard to check that if the conditions hold, the positive roots of qQ are
in turn in bijection with the positive roots of the root system associated with
, via the map taking a positive root v of qQ to the element
vi αi
i∈Q0
It is positive definite and its positive roots are indeed precisely the dimension
vectors
V : V1
f
/ V2
R → 0 , R2 → C , R → C , 0 → C.
1= 1-< 1-=
=== ---<<< ---===
== -- << -- ==
= < =
2 3 4 2 3 4 5 2 3 4 5 6.
Let us recall the list of simply laced extended Dynkin quivers. In each case,
the number of vertices of the diagram D n equals n + 1.
n : ◦ SSSS
A kkk SSS
kkkk . . .
◦ ◦ ◦ ◦
n :
D ◦ EE ◦
E yyy
◦ ◦ ... ◦ ◦ EE
yyy E
◦ ◦
6 :
E ◦ ◦ ◦ ◦ ◦
◦
◦
7 :
E ◦ ◦ ◦ ◦ ◦ ◦ ◦
◦
8 : ◦
E ◦ ◦ ◦ ◦ ◦ ◦ ◦
◦
without oriented cycles and let ν(Q) be the supremum of the multiplicities of
the arrows occurring in all quivers mutation-equivalent to Q.
Theorem 5.5.
a) Q is representation-finite iff ν(Q) equals 1.
b) Q is tame iff ν(Q) equals 2.
c) Q is wild iff ν(Q) ≥ 3 iff ν(Q) = ∞.
d) The mutation class of Q is finite iff Q has two vertices, is representation-
finite or tame.
Here, part a) follows from Gabriel’s theorem and part (iii) of Theorem 1.8
in [52]. Part b) follows from parts a) and c) by exclusion of the third. For part
c), let us first assume that Q is wild. Then it is proved at the end of the proof of
theorem 3.1 in [13] that ν(Q) = ∞. Conversely, let us assume that ν(Q) ≥ 3.
Then using Theorem 5 of [29] we obtain that Q is wild. Part d) is proved in
[13].
of the corresponding root system, there is a unique non initial cluster variable
Xα with denominator
x1d1 . . . xndn .
By combining this with Gabriel’s theorem, we get the
Corollary 5.6. The map taking an indecomposable representation V with
dimension vector (di ) of Q to the unique non initial cluster variable XV whose
denominator is x1d1 . . . xndn induces a bijection from the set of ismorphism classes
of indecomposable representations to the set of non initial cluster variables.
Let us consider this bijection for Q = A2 :
S2 = (0 → k) P1 = (k → k) S1 = (k → 0)
1 + x1 x1 + 1 + x2 1 + x2
XS2 = XP1 = SS1 =
x2 x1 x2 x1
Cluster algebras, representations, triangulated categories 107
We observe that for the two simple representations, the numerator contains
exactly two terms: the number of subrepresentations of the simple represen-
tation! Moreover, the representation P1 has exactly three subrepresentations
and the numerator of XP1 contains three terms. In fact, it turns out that this
phenomenon is general in type A. But now let us consider the following quiver
with underlying graph D4
3 NNN
NNN
'/
2
ppp7 4
ppp
1
Here the sum is taken over all vectors e ∈ Nn such that 0 ≤ ei ≤ di for all i. For
each such vector e, the quiver Grassmannian Gre (V ) is the variety of n-tuples of
subspaces Ui ⊂ Vi such that dim Ui = ei and the Ui form a subrepresentation
of V . By taking such a subrepresentation to the family of the Ui , we obtain a
108 Bernhard Keller
map
n
Gre (V ) → Grei (Vi ) ,
i=1
Q, we have the lazy path ei = (i|i), the unique path of length 0 which starts
at i and stops at i and does nothing in between. The path category has set of
objects Q0 (the set of vertices of Q) and, for any vertices i, j , the morphism
space from i to j is the vector space whose basis consists of all paths from i
to j . Composition is induced by composition of paths and the unit morphisms
are the lazy paths. If Q is finite, we define the path algebra to be the matrix
algebra
kQ = Hom(i, j )
i,j ∈Q0
Pi = ei kQ.
endowed with the natural right action of the path algebra. Conversely, a kQ-
module M gives rise to the representation V with Vi = Mei for each vertex i
of Q and Vα given by right multiplication by α for each arrow α of Q. The
category mod kQ is abelian, i.e. it is additive, has kernels and cokernels and
for each morphism f the cokernel of its kernel is canonically isomorphic to the
kernel of its cokernel.
The category mod kQ is hereditary. Recall from [32] that this means that
submodules of projective modules are projective; equivalently, that all extension
110 Bernhard Keller
for all i ∈ Z, where M[i] denotes the complex M shifted by i degrees to the left:
M[i]p = M p+i , p ∈ Z, and endowed with the differential dM[i] = (−1)i dM .
The category DQ has all finite direct sums (and they are given by direct sums of
complexes) and the decomposition theorem 5.1 holds. Moreover, each object
is isomorphic to a direct sum of shifted copies of modules (this holds more
generally in the derived category of any hereditary abelian category, for example
the derived category of coherent sheaves on an algebraic curve). The category
DQ is abelian if and only if the quiver Q does not have any arrows. However,
it is always triangulated. This means that it is k-linear (it is additive, and
the morphism sets are endowed with k-vector space structures so that the
composition is bilinear) and endowed with the following extra structure:
a) a suspension (or shift) functor : DQ → DQ , namely the functor taking a
complex M to M[1];
Cluster algebras, representations, triangulated categories 111
= u1 B (5.5.1)
σ (α) zzz u BBBα
z
zz oo7
2 NN B
NNNBB
zozooo . N'
τ v OO 7v
OOO .. pppp
O' pp
us
112 Bernhard Keller
ind DA5
(1,5) • •
(1,4) • • ΣP2 • •
(1,3) • P3 • • • • •
(1,2) • P2 • • • •Σ 2
P2 • •
(1,1) • (2,1) • • • • • • •
where v runs through the vertices of ZQ. The mesh category is the quotient of
the path category of ZQ by the mesh ideal.
In figure 1, we see the repetition for Q = A5 and the map taking its vertices
to the indecomposable objects of the derived category. The vertices marked •
belonging to the left triangle are mapped to indecomposable modules. The
vertex (1, i) corresponds to the indecomposable projective Pi . The arrow
(1, i) → (1, i + 1), 1 ≤ i ≤ 5, is mapped to the left multiplication by the arrow
i → i + 1. The functor takes a mesh (5.5.1) to a triangle
Mτ v / s Mu i / Mv / Mτ v (5.5.2)
i=1
0 / Mτ v / s Mu i / Mv /0
i=1
Cluster algebras, representations, triangulated categories 113
[X] − [Y ] + [Z]
Using these equalities, we can easily determine dim M for each indecomposable
M starting from the known dimension vectors dim Pi , 1 ≤ i ≤ n. In the above
example, we find the dimension vectors listed in figure (2).
Thanks to the theorem, the automorphism τ of the repetition yields a k-
linear automorphism, still denoted by τ , of the derived category DQ . This
automorphism has several intrinsic descriptions:
1) As shown in [60], it is the right derived functor of the left exact Coxeter
functor rep(Qop ) → rep(Qop ) introduced by Bernstein-Gelfand-Ponomarev
[19] in their proof of Gabriel’s theorem. If we identify K0 (DQ ) with the root
114 Bernhard Keller
lattice via Gabriel’s theorem, then the automorphism induced by τ −1 equals the
the Coxeter transformation c. As shown by Gabriel [60], the identity ch = 1,
where h is the Coxeter number, lifts to an isomorphism of functors
τ −h → 2 .
∼
(5.5.3)
2) It can be expressed in terms of the Serre functor of DQ : Recall that for
a k-linear triangulated category T with finite-dimensional morphism spaces,
a Serre functor is an autoequivalence S : T → T such that the Serre duality
formula holds: We have bifunctorial isomorphisms
∼
D Hom(X, Y ) → Hom(Y, SX) , X, Y ∈ T ,
where D is the duality Homk (?, k) over the ground field. Notice that this
determines the functor S uniquely up to isomorphism. In the case of DQ =
Db (mod kQ), it is not hard to prove that a Serre functor exists (it is given by
the left derived functor of the tensor product by the bimodule D(kQ)). Now
the autoequivalence τ , the suspension functor and the Serre functor S are
linked by the fundamental isomorphism
∼
τ → S. (5.5.4)
0→L→E →M →0,
we have
CC(L)CC(M) = CC(E).
b2) If
0→L→E→M→0
CC(E) + 1 = CC(L)CC(M).
is the orbit category of the derived category under the action of the cyclic group
generated by the autoequivalence τ −1 = S −1 2 . This means that the objects
of CQ are the same as those of the derived category DQ and that for two objects
X and Y , the morphism space from X to Y in CQ is
CQ (X, Y ) = DQ (X, (S −1 2 )p Y ).
p∈Z
Morphisms are composed in the natural way. This definition is due to Buan-
Marsh-Reineke-Reiten-Todorov [10], who were trying to obtain a better under-
standing of the ‘decorated quiver representations’ introduced by Reineke-
Marsh-Zelevinsky [106]. For quivers of type A, an equivalent category was
defined independently by Caldero-Chapoton-Schiffler [28] using an entirely
116 Bernhard Keller
Ext 1 (Ti , Tj ) = 0
for all i, j .
c) If T1 , . . . , Tn is cluster-tilting, then the quiver (cf. below) of the endomor-
phism algebra of the sum T = ni=1 Ti does not have loops nor 2-cycles and
Cluster algebras, representations, triangulated categories 117
where radT (Ti , Tj ) denotes the vector space of non isomorphisms from Ti to
Tj (thanks to the locality of the endomorphism rings, this set is indeed closed
under addition) and rad2T the subspace of non isomorphisms admitting a non
trivial factorization:
n
rad2T (Ti , Tj ) = radT (Tr , Tj ) radT (Ti , Tr ).
r=1
T3
? ???
4 20
? ???? ? ???
? ?
?3? ?8? ? 13 ? ? 21 ?
??? ??? ??? ???
? 2 ?? T
? ??2 12
? ? 16
? ? ? 22 ?
?? ?? ??? ??? ???
? 1 ?? ? 6 ?? ?11 ?? ? 15 ? ? 18 ?
? ? 23 ?
?? ?? ?? ??? ?? ???
0 T1 10 T4 17 T5 24
obtain the orbit quiver ZQ/(τ −1 )Z . In the orbit category, we have τ → so
∼
that T1 is the indecomposable associated to vertex 0, for example. Using this
and the description of the morphisms in the mesh category, it is easy to check
that we do have
Ext1 (Ti , Tj ) = 0
HomCQ (Ti , Tj )
This algebra is easily seen to be isomorphic to the algebra given by the following
quiver Q
5 =o
γ
3
== @
== α
β
@ 2 == β
α
==
=
1 o 4
γ
αβ = 0 , βγ = 0 , γ α = 0.
Cluster algebras, representations, triangulated categories 119
Thus the quiver of End(T ) is Q . It encodes the exchange matrix of the associ-
ated cluster
1 + x2
XT1 =
x1
x1 x2 + x1 x4 + x3 x4 + x2 x3 x4
XT2 =
x1 x2 x3
x1 x2 x3 + x1 x2 x3 x4 + x1 x2 x5 + x1 x4 x5 + x3 x4 x5 + x2 x3 x4 x5
XT3 =
x1 x2 x3 x4 x5
x2 + x4
XT4 =
x3
1 + x4
XT5 = .
x5
TR2 A sequence
u v w
X → Y → Z → X
120 Bernhard Keller
and
v
∗
Tk → Tj → Tk → ∗Tk ,
arrows
j →k
XN = XN1 XN2
By combining this with the exchange triangles, we see that in the situation of
c), we have
XL XM = XE + XE .
We would like to generalize this identity to the case where the space Ext1 (L, M)
is of higher dimension. For three objects L, M and N of CQ , let Ext 1 (L, M)N
be the subset of Ext1 (L, M) formed by those morphisms ε : L → M such
that in the triangle
ε
M → E → L → M,
Theorem 5.14 ([30]). Suppose that L and M are objects of CQ such that
Ext1 (L, M)
= 0. Then we have
χ (P Ext1 (L, M)N ) + χ (P Ext1 (M, L)N )
XL XM = XN ,
N
χ (P Ext1 (L, M))
where the brackets denote isomorphism classes and the vertical bars the cardi-
nalities of the underlying sets, cf. Proposition 1.5 of [119].
where CC(M) is defined as in section 5.3 Notice that in general, XL can only
be expected to be an element of the fraction field Q(x1 , . . . , xn ), not of the
cluster algebra AQ inside this field. (The exponents in the formula for XL are
perhaps more transparent in equation 7.5.1 below).
Theorem 6.1. Let Q be a finite quiver without oriented cycles with vertex set
{1, . . . , n}.
a) The map L
→ XL induces a bijection from the set of isomorphism classes
of rigid indecomposables of the cluster category CQ onto the set of cluster
variables of the cluster algebra AQ .
b) Under this bijection, the clusters correspond exactly to the cluster-tilting
sets, i.e. the sets T1 , . . . , Tn of rigid indecomposables such that
Ext 1 (Ti , Tj ) = 0
for all i, j .
c) For a cluster-tilting set T1 , . . . , Tn , the quiver of the endomorphism algebra
of the sum of the Ti does not have loops nor 2-cycles and encodes the
exchange matrix of the [68] seed containing the corresponding cluster.
d) If L and M are rigid indecomposables such that the space Ext1 (L, M) is
one-dimensional, then we have the generalized exchange relation
XL XM = XB + XB (6.1.1)
where B and B are the middle terms of ‘the’ non split triangles
L /B /M / L and M / B /L / M .
Parts a), b) and d) of the theorem are proved in [29] and part c) in
[11]. The proofs build on work by many authors notably Buan-Marsh-
Reiten-Todorov [12] Buan-Marsh-Reiten [11], Buan-Marsh-Reineke-Reiten-
Todorov [10], Marsh-Reineke-Zelevinsky [106], . . . and especially on Caldero-
Chapoton’s explicit formula for XL proved in [27] for orientations of simply
laced Dynkin diagrams. Another crucial ingredient of the proof is the Calabi-
Yau property of the cluster category. An alternative proof of part c) was given
by A. Hubery [79] for quivers whose underlying graph is an extended simply
laced Dynkin diagram.
We describe the main steps of the proof of a). The mutation of cluster-tilting
sets is defined using the construction of section 5.10.
XT1 XT1∗ = XE + XE .
It follows from 1)-3) that the map L → XL does take rigid indecomposables to
cluster variables and that each cluster variable is obtained in this way. It remains
to be shown that a rigid indecomposable L is determined up to isomorphism
by XL . This follows from
Conjecture 6.2. Suppose that Q does not have oriented cycles. Then all cluster
variables of AQ belong to N[x1± , . . . , xn± ].
For acyclic cluster algebras, parts a) and b) confirm conjecture 4.14 parts
(3) and (4) by Fomin-Zelevinsky in [53]. By b), the cluster algebra associated
with a quiver without oriented cycles has a well-defined cluster-type.
is the quiver
For example, if
1
α /2 β
/3,
and the ideal generated by the above sum of commutators is also generated by
the elements
α ∗ α, αα ∗ − β ∗ β, ββ ∗ .
It is classical, cf. e.g. [115], that the algebra = () is a finite-dimensional
(!) selfinjective algebra (i.e. is injective as a right -module over itself). Let
mod denote the category of k-finite-dimensional right -modules. The stable
module category mod is the quotient of mod by the ideal of all morphisms
factoring through a projective module. This category carries a canonical trian-
gulated structure (like any stable module category of a self-injective algebra):
The suspension is constructed by choosing exact sequences of modules
0 → L → I L → L → 0
where I L is injective (but not necessarily functorial in L; the object L
becomes functorial in L when we pass to the stable category). The triangles are
by definition isomorphic to standard triangles obtained from exact sequences
of modules as follows: Let
i p
0→L→M→N →0
128 Bernhard Keller
0 /L i /M p
/N /0
1 e
0 /L / IL / L /0
Then the image of (i, p, e) is a standard triangle in the stable module category.
As shown in [36], the stable module category C = mod is 2-Calabi-Yau and
it is easy to check that assumption 1) holds.
Theorem 7.1 (Geiss-Leclerc-Schröer). The category C = mod admits a
cluster-tilting object T such that the quiver of End(T ) is obtained from that of
the category of indecomposable k-modules by deleting the injective vertices
and adding an arrow v → τ v for each non projective vertex v.
Here, by the quiver of the category of indecomposable k-modules, we
mean the full subquiver of the repetition ZQ which is formed by the vertices
corresponding to modules (complexes concentrated in degree 0). Thus for
= A5 , this quiver is as follows:
P5? = ?I1
???
? P4 ?? ? I2 ???
?? ?
P3 ?7? ? I3 ???
? ??? ??? ?
?
P ? ?
6 10
? ? ? I4 ???
? ???
2
? ??? ??? ?
P1 5 9 12 I5
where we have marked the indecomposable projectives Pi and the indecom-
posable injectives Ij . If we remove the vertices corresponding to the indecom-
posable injectives (and all the arrows incident with them) and add an arrow
v → τ v for each vertex not corresponding to an indecomposable projective,
we obtain the following quiver
?0?
???
o
? 1 ?? ?2?
?? ???
o o
? 3 ?? ? 4 ?? ? ???
5
?? ?? ?
6o 7o 8o 9
Cluster algebras, representations, triangulated categories 129
In a series of papers [64] [66] [65] [62] [61] [67] [63], Geiss-Leclerc-Schröer
have obtained remarkable results for a class of quivers which are important
in the study of (dual semi-)canonical bases. They use an analogue [67] of
the Caldero-Chapoton map due ultimately to Lusztig [104]. The class they
consider has been further enlarged by Buan-Iyama-Reiten-Scott [7]. Thanks
to their results, an analogue of Caldero-Chapoton’s formula and a weakened
version of theorem 6.1 was proved in [57] for an even larger class.
C = U/Tj |j ∈ J
7.3. Mutation
Let (C, T ) be a 2-Calabi-Yau category with cluster-tilting object. Let T1 be an
indecomposable direct factor of T .
endomorphism algebra of T may contain loops and 2-cycles and then the quiver
of the endomorphism algebra of μ1 (T ) is not determined by Q. Let us illustrate
this phenomenon on the following example (taken from Proposition 2.6 of
[25]): Let C be the orbit category of the bounded derived category DD 6 under
the action of the autoequivalence τ 2 . Then C satisfies the assumptions 1) and
2) of section 7.1. Its category of indecomposables is equivalent to the mesh
category of the quiver
C? C ?C?
?? ? ???? ???
?
/ 5 / B / 11 / B / 17
B ? ?? ? ?? ?
?? ??
1?
? ???
7 13
?? ? ???
? ? ?
?4? ? 10 ? ? 16
??? ???
A A A
where the vertices labeled A, 1, B, C, 4, 5 on the left have to be identified
with the vertices labeled A, 13, B, C, 16, 17 on the right. In this case, there
are exactly 6 indecomposable rigid objects, namely A, B, C, A , B and C .
There are exactly 6 cluster-tilting sets. The following is the exchange graph: Its
vertices are cluster-tilting sets (we write AC instead of {A, C}) and its edges
represent mutations.
AC AB6
66
66
6
B C6 CB
66
66
6
B A C A
The quivers of the endomorphism algebras are as follows:
◦o /◦
AC, AB : d
B C, C B : ◦o /◦
/◦
B A ,C A : :◦o
In the setting of the above theorem, there are still exchange triangles as in
theorem 5.12 but their description is different: Let T be the full subcategory
of C formed by the direct sums of indecomposables Ti , where i is different
from k. A left T -approximation of Tk is a morphism f : Tk → T with T
Cluster algebras, representations, triangulated categories 131
Theorem 7.4 (Iyama-Yoshino [81]). If (T1 , T1∗ ) is an exchange pair, there are
non split triangles, unique up to isomorphism,
f g
T1 → E → T1∗ → T1 and T1∗ → E → T1 → T1∗
T1∗ → E → T1 → T1∗
the exchange triangle. The long exact sequence induced in C(T1 , ?) by this
triangle yields a short exact sequence
where the leftmost term is the space of those endomorphisms of T1 which factor
through a sum of copies of T /T1 . Now the algebra C(T1 , T1 ) is local and its
residue field is k (since k is algebraically closed). We deduce the following
lemma.
Lemma 7.5. The quiver of the endomorphism algebra of T does not have
a loop at the vertex corresponding to T1 iff we have dim Ext1 (T1 , T1∗ ) = 1 iff
Ext1 (T1 , T1∗ ) is a simple module over C(T1 , T1 ). In this case, in the exchange
triangles
we have
E= Ti and E = Tj .
arrows arrows
i→1 1→j
such that the quiver of End(T (i) ) does not have loops nor 2-cycles for all
1 ≤ i ≤ N . We define a rigid indecomposable of C to be reachable from T if it
is a direct summand of a reachable cluster-tilting object.
T1 → T0 → L → T1
[T0 ] − [T1 ]
In the situation of the lemma, we define the index ind(L) of L as the element
[T0 ] − [T1 ] of K0 (T ).
n
n
Si ,ea
XL =
g (L)
xi i χ (Gre (Ext1 (T , L))) xi , (7.5.1)
i=1 e i=1
Theorem 7.11 (Palu [112]). If L and M are objects of C such that Ext1 (L, M)
is one-dimensional, then we have
XL XM
= XE + XE ,
where
L → E → M → L and M → E → L → M
are ‘the’ two non split triangles. Thus, if L is a rigid indecomposable reachable
from T , then XL is a cluster variable of AQ .
x = XL (u1 , . . . , un ).
The expression for XL makes it natural to define the polynomial FL ∈
Z[y1 , . . . yn ] by
n
FL =
e
χ (Gre (Ext1 (T , L))) yj j . (7.5.2)
e j =1
We then have
! n "
n
n
XL xi i FL
g (L)
= xibi1 , . . . , xibin .
i=1 i=1 i=1
where the cli are the exponents in the tropical Y -variables yi,t = nl=1 ylcli and
B = (bij ) is the antisymmetric matrix associated with Qt .
Finally, the non tropical Y -variables Yj,t lie in the field Q(y1 , . . . , yn ). At
t = t0 , we have Yj,t = yj , 1 ≤ j ≤ n, and if t and t are linked by an edge
136 Bernhard Keller
labeled i, then
−1
Yi,t if j = i
Yj,t = [bij ]+ −bij
(7.6.1)
Yj,t Yi,t (Yi,t + 1) if j
= i.
U/Ti |i ∈ Q0
be a path in Tn and suppose that for each 1 ≤ i ≤ N , the quiver of the endo-
morphism algebra of Tti does not have loops nor 2-cycles. Let t = tN and let
T1 , . . . , Tn be the indecomposable summands of T = Tt . Recall from part b)
of theorem 7.9 that the indices ind(Tl ), 1 ≤ l ≤ n, form a basis of the group
K0 (T ), where T is the full subcategory of C formed by the direct summands
of finite direct sums of copies of T .
Theorem 7.13.
a) The exchange matrix Bt = (bij ) associated with t is the antisymmetric
matrix associated with the quiver of the endomorphism algebra of Tt .
c
b) We have yj,t = nl=1 yl lj , 1 ≤ j ≤ n, where clj is defined by
n
[Tj ] = clj ind(Tl ).
l=1
DA /(S −1 2 )Z .
We consider two classes of examples obtained from this theorem: First, let
be a simply laced Dynkin diagram and k the path algebra of a quiver with
i.e. the endomorphism
underlying graph . Let A the Auslander algebra of k ,
algebra of the direct sum of a system of representatives of the indecompos-
able B-modules modulo isomorphism. Then it is not hard to check that the
138 Bernhard Keller
•o ◦ /•o
@◦
@ ^===
===
• 5o / o
55 {= ◦ 5iR5R RR =•==55 {= ◦ 55
5 {{{{ 55 RRR =5=5 {{{
RRR {
55
o
◦O NN ◦O / o
◦O NN ◦
NNN
NNN ppppp
pp NNNN O
& xp N N&
•o p 8 ◦ fNNN /•o
p8 ◦
pp p NNN pp p
pp p p NN pp pp
◦o ◦ /◦o ◦
◦J /◦o ◦J /◦
◦ X1 / ◦ 1o
◦ X1 / ◦ 1
11 11 11 11
1 1
1 1
◦o ◦O /◦o ◦O
◦O /◦o ◦O /◦
◦o ◦ /◦o ◦
5
Figure 5. The quiver A4 D
the identity matrix of the same format, then A = 2J − C. Let h and h be the
Coxeter numbers of and .
The Y -system of algebraic equations associated with the pair of Dynkin
diagrams (, ) is a system of countably many recurrence relations in the
variables Yi,i ,t , where (i, i ) is a vertex of × and t an integer. The system
reads as follows:
j ∈I (1 + Yj,i ,t )
aij
Yi,i ,t−1 Yi,i ,t+1 = . (8.0.1)
−1 ai j
j ∈I (1 + Yi,j ,t )
Yi,i , (i, i ) ∈ I × I . We can apply mutations to it using the quiver mutation rule
and the mutation rule for (non tropical) Y -variables given in equation 7.6.1.
A general construction. Let R be a quiver and v a sequence of vertices
v1 , . . . , vN of R. We assume that the composed mutation
μv = μvN . . . μv2 μv1
transforms R into itself. Then clearly the same holds for the inverse sequence
μ−1
v = μv1 μv2 . . . μvN .
Notice that there are no arrows between any two vertices of the index set so that
the order in the product does not matter. Then it is easy to check that μ+,+ μ−,−
and μ−,+ μ+,− both transform QQ into (QQ )op and vice versa. Thus the
composed sequence of mutations
μ = μ−,− μ+,+ μ−,+ μ+,−
transforms QQ into itself. We define the Y -system y associated with QQ
to be the restricted Y -pattern associated with QQ and μ . As in section 8
of [55], one checks that the identity ϕ h+h = 1 follows if one shows that the
system y is periodic of period dividing h + h .
Third step. One checks easily that we have
μ+− (QQ ) = Q ⊗ Q ,
where the tensor product Q ⊗ Q is defined at the end of section 7.7. For the
above quivers A4 and D 5 , the tensor product is depicted in figure 4. Therefore,
the periodicity of the restricted Y -system associated with QQ and μ is
142 Bernhard Keller
Tt = μt⊗ (T )
The proposition is proved by showing that the indices of the two objects are
equal. This suffices by theorem 7.9. Now we conclude thanks to the following
categorical periodicity result:
h+h
Proposition 8.5. The power is isomorphic to the identity functor.
Let us sketch the proof of this proposition: With the natural abuse of notation,
the Serre functor S of the bounded derived category of kQ ⊗ kQ is given by the
‘tensor product’ S ⊗ S of the Serre functors for kQ and kQ . In the generalized
cluster category, the Serre functor becomes isomorphic to the square of the
suspension functor. So we have
S = S ⊗ S = 2 = ⊗
Cluster algebras, representations, triangulated categories 143
τ ⊗τ =1
τ −1 ⊗ 1 = 1 ⊗ τ.
Now we compute h+h by using the left hand side for the first h factors and
the right hand side for the last h factors:
h+h
= (τ −1 ⊗ 1)h (1 ⊗ τ )h .
h+h
= ( 2 ⊗ 1)(1 ⊗ −2 ) = 2 −2 = 1
as required.
The module categories over kQ and kQ are linked by a pair of adjoint
functors [19]
Mod kQ
O
F0 G0
Mod kQ.
In fact, the sum of the images of all the Pj has a natural structure of complex of
kQ -kQ-bimodules and F can also be described as the derived tensor product
over kQ with this complex of bimodules.
The example of the following quiver Q
1
E 222 a
b 22
2o c 3
Q : 2 ← 1 ← 3
shows that if we mutate at a vertex which is neither a sink nor a source, then
the derived categories of representations of Q and Q are not equivalent in
Cluster algebras, representations, triangulated categories 145
#
∂a : HH0 → kQ
Q
→ μi (Q)
(Q, W )
→ μi (Q, W ) = (Q , W ) ,
1
Q: E 222 a
b (9.2.1)
22
2o c 3
with the potential W = abc at the vertex 1 is the quiver with potential
Q : 2 ← 1 ← 3 , W = 0.
On the other hand, the mutation of the above cyclic quiver Q with the potential
W = abcabc at the vertex 1 is the quiver
1 ^=
b ===a
==
=
o
e /
2 3
c
# α W | α ∈ Q1 ).
P(Q, W ) = kQ/(∂
Let us consider two typical examples: Let Q be the quiver with one ver-
tex and three loops labeled X, Y and Z. Let W = XY Z − XZY . Then the
cyclic derivatives of W yield the commutativity relations between X, Y and Z
and the Jacobi algebra is canonically isomorphic to the power series algebra
k[[X, Y, Z]]. It is not hard to check that in this example, the homology of is
concentrated in degree 0 so that we have a quasi-isomorphism → P(Q, W ).
Using theorem 5.3.1 of Ginzburg’s [71], one can show that this is the case if and
only if the full subcategory of the derived category of the Jacobi algebra formed
by the complexes whose homology is of finite total dimension is 3-Calabi-Yau
as a triangulated category (cf. also below).
As a second example, we consider the Ginzburg dg algebra associated
with the cyclic quiver 9.2.1 with the potential W = abc. Here is the graded
148 Bernhard Keller
quiver Q:
t2
@ 2 =^==== ∗
b∗ ====a
==
a ===
b
:1o
c
t1 /3d t3
c∗
H i () = CA3 (T , i T )
where A3 is the quiver 1 → 2 → 3, CA3 its cluster category and T the sum of the
images of the modules P1 , P3 and P3 /P2 . Since is an autoequivalence of finite
order of the cluster category CA3 , we see that has non vanishing homology
in infinitely many degrees < 0. In particular, is not quasi-isomorphic to the
Jacobi algebra.
Let us denote by D the derived category of . Its objects are the differential
Z-graded right -modules and its morphisms obtained from morphisms of dg
-modules (homogeneous of degree 0 and which commute with the differential)
by formally inverting all quasi-isomorphisms, cf. [89]. Let us denote by per
the perfect derived category, i.e. the full subcategory of D which is the closure
of the free right -module . under shifts, extensions and passage to direct
factors. Finally, we denote by Df d () the finite-dimensional derived category,
i.e. the full subcategory of D formed by the dg modules whose homology
is of finite total dimension (!). We recall that the objects of per() can be
intrinsically characterized as the compact ones, i.e. those whose associated
covariant Hom-functor commutes with arbitrary coproducts. The objects M of
the bounded derived category are characterized by the fact that Hom(P , M) is
finite-dimensional for each object P of per().
The following facts will be shown in [85]. Notice that they hold for arbitrary
Q and W .
1) The dg algebra is homologically smooth, i.e. it is perfect as an object
of the derived category of e = ⊗ op . This implies that the bounded derived
category is contained in the perfect derived category, cf. e.g. [88].
Cluster algebras, representations, triangulated categories 149
The name is justified by the fact that if Q is a quiver without oriented cycles
(and so W = 0), then C(Q,W ) is triangle equivalent to CQ . In general, the cate-
gory C(Q,W ) has infinite-dimensional Hom-spaces. However, if H 0 () is finite-
dimensional, then C(Q,W ) is a 2-Calabi-Yau category with cluster-tilting object
T = , in the sense of section 7.1, cf. [2] (the present version of [2] uses non
completed Ginzburg algebras; the complete case will be included in a future
version).
From now on, we suppose that W does not involve cycles of length ≤ 1.
Then the simple Q-modules Si associated with the vertices of Q yield a basis
of the Grothendieck group K0 (Df d ()). Thanks to the Euler form
P , M = χ (RHom(P , M)) , P ∈ per() , M ∈ Df d () ,
this Grothendieck group is dual to K0 (per()) and the dg modules Pi = ei
form the basis dual to the Si . The Euler form also induces an antisymmetric
bilinear form on K0 (Df d () (‘Poisson form’). If W does not involve cycles of
length ≤ 2, then the quiver of the Jacobi algebra is Q and the matrix of the
form on Df d () in the basis of the Si is given by
Si , Sj = |{arrows j → i of Q}| − |{arrows i → j of Q}.
The dual of this form is given by the map (‘symplectic form’)
op ) = K0 (per()) ⊗Z K0 (per())
K0 (k) → K0 (per( ⊗
associated with the k- e -bimodule .
Now suppose that W does not involve cycles of length ≤ 2 and that Q does
not have loops nor 2-cycles. Then each simple Si is a (3-)spherical object in the
150 Bernhard Keller
Moreover, the quiver Q encodes the dimensions of the Ext1 -groups between
the Si : We have
dim Ext1 (Si , Sj ) = |{arrows j → i in Q}|.
In particular, for i
= j , we have either Ext1 (Si , Sj ) = 0 or Ext1 (Sj , Si ) = 0
because Q does not contain 2-cycles. It follows that the Si , i ∈ Q0 , form a
spherical collection in Df d () in the sense of [98]. Conversely, each spherical
collection in an (A∞ -) 3-Calabi-Yau category in the sense of [98] can be
obtained in this way from the Ginzburg algebra associated to a quiver with
potential.
The categories we have considered so far are summed up in the sequence of
triangulated categories
In fact, the functor F is given by the left derived functor of the tensor product
by a suitable --bimodule.
By transport of the canonical t-structure on D( ), we obtain new t-structures
on D() and Df d (). They are related to the canonical ones by a tilt (in the
Cluster algebras, representations, triangulated categories 151
@ @ 2 =
========
@ ======
Q: =====
=
0 oo 1
o
where the arrows going out from i are labeled xi , yi , zi , 0 ≤ i ≤ 2, endowed
with the potential
2
W = (xi yi zi − xi zi yi ).
i=0
Db (coh(ω))/DZb (coh(ω))
and thus with the bounded derived category Db (coh(ω \ Z)) of coherent
sheaves on the complement of the zero section of ω. In order to understand
why this category is ‘close to being’ Calabi-Yau of dimension 2, we consider
152 Bernhard Keller
Recall that a tilting module over an algebra B is a B-module T such that the
total derived functor of the tensor product by T over the endomorphism algebra
EndB (T ) is an equivalence
∼
D(EndB (T )) → D(B).
The second assertion of part a) of the following theorem generalizes a result
by Assem-Brüstle-Schiffler [3].
Cluster algebras, representations, triangulated categories 153
References
[1] Claire Amiot, Cluster categories for algebras of global dimension 2 and quivers
with potential, Annales de l’institut Fourier 59 (2009), no. 6, 2525–2590.
[2] Claire Amiot, Sur les petites catégories triangulées, Ph. D. thesis, Université
Paris Diderot - Paris 7, Juillet 2008.
[3] Ibrahim Assem, Thomas Brüstle, and Ralf Schiffler, Cluster-tilted algebras as
trivial extensions, Bull. London Math. Soc. 40 (2008), 151–162.
[4] Ibrahim Assem, Daniel Simson, and Andrzej Skowroński, Elements of the rep-
resentation theory of associative algebras. Vol. 1, London Mathematical Society
Student Texts, vol. 65, Cambridge University Press, Cambridge, 2006, Tech-
niques of representation theory.
[5] M. Auslander, I. Reiten, and S. Smalø, Representation theory of Artin algebras,
Cambridge Studies in Advanced Mathematics, vol. 36, Cambridge University
Press, 1995 (English).
[6] Maurice Auslander, Functors and morphisms determined by objects, Represen-
tation theory of algebras (Proc. Conf., Temple Univ., Philadelphia, Pa., 1976),
Dekker, New York, 1978, pp. 1–244. Lecture Notes in Pure Appl. Math., Vol. 37.
[7] Aslak Bakke Buan, Osamu Iyama, Idun Reiten, and Jeanne Scott,
Cluster structures for 2-Calabi-Yau categories and unipotent groups,
arXiv:math.RT/0701557. Compos. Math. 145 (2009), no. 4, 1035–1079.
[8] Aslak Bakke Buan, Osamu Iyama, Idun Reiten, and David Smith, Mutation of
cluster-tilting objects and potentials, arXiv:0804.3813. to appear in Amer. J.
Math.
[9] Aslak Bakke Buan and Robert Marsh, Cluster-tilting theory, Trends in represen-
tation theory of algebras and related topics, Contemp. Math., vol. 406, Amer.
Math. Soc., Providence, RI, 2006, pp. 1–30.
[10] Aslak Bakke Buan, Robert J. Marsh, Markus Reineke, Idun Reiten, and Gordana
Todorov, Tilting theory and cluster combinatorics, Advances in Mathematics 204
(2) (2006), 572–618.
[11] Aslak Bakke Buan, Robert J. Marsh, and Idun Reiten, Cluster mutation via quiver
representations, Comm. Math. Helv. 83 (2008), 143–177.
[12] Aslak Bakke Buan, Robert J. Marsh, Idun Reiten, and Gordana Todorov, Clusters
and seeds in acyclic cluster algebras, Proc. Amer. Math. Soc. 135 (2007), no. 10,
3049–3060 (electronic), With an appendix coauthored in addition by P. Caldero
and B. Keller.
[13] Aslak Bakke Buan and Idun Reiten, Acyclic quivers of finite mutation type, Int.
Math. Res. Not. (2006), Art. ID 12804, 10.
[14] Michael Barot, Christof Geiss, and Andrei Zelevinsky, Cluster algebras of finite
type and positive symmetrizable matrices, J. London Math. Soc. (2) 73 (2006),
no. 3, 545–564.
Cluster algebras, representations, triangulated categories 155
[15] Andre Beineke, Thomas Bruestle, and Lutz Hille, Cluster-cyclic quivers with
three vertices and the Markov equation, arXiv:math/0612213. to appear in Algebr.
Represent. Theory.
[16] Arkady Berenstein, Sergey Fomin, and Andrei Zelevinsky, Parametrizations of
canonical bases and totally positive matrices, Adv. Math. 122 (1996), no. 1,
49–149.
[17] , Cluster algebras. III. Upper bounds and double Bruhat cells, Duke
Math. J. 126 (2005), no. 1, 1–52.
[18] David Berenstein and Michael Douglas, Seiberg duality for quiver gauge theories,
arXiv:hep-th/0207027v1.
[19] I. N. Bernšteı̆n, I. M. Gel fand, and V. A. Ponomarev, Coxeter functors, and
Gabriel’s theorem, Uspehi Mat. Nauk 28 (1973), no. 2(170), 19–33.
[20] Klaus Bongartz, Algebras and quadratic forms, J. London Math. Soc. (2) 28
(1983), no. 3, 461–469.
[21] Tom Bridgeland, Stability conditions and Hall algebras, Talk at the meeting
‘Recent developments in Hall algebras’, Luminy, November 2006.
[22] , t-structures on some local Calabi-Yau varieties, J. Algebra 289 (2005),
no. 2, 453–483.
[23] , Derived categories of coherent sheaves, International Congress of Math-
ematicians. Vol. II, Eur. Math. Soc., Zürich, 2006, pp. 563–582.
[24] , Stability conditions on triangulated categories, Ann. of Math. (2) 166
(2007), no. 2, 317–345.
[25] Igor Burban, Osamu Iyama, Bernhard Keller, and Idun Reiten, Cluster tilting
for one-dimensional hypersurface singularities, Adv. Math. 217 (2008), no. 6,
2443–2484.
[26] F. Cachazo, B. Fiol, K. Intriligator, S. Katz, and C. Vafa, A geometric unification
of dualities, Nucl. Phys. B 628 (2002), 3–78.
[27] Philippe Caldero and Frédéric Chapoton, Cluster algebras as Hall algebras of
quiver representations, Comment. Math. Helv. 81 (2006), no. 3, 595–616.
[28] Philippe Caldero, Frédéric Chapoton, and Ralf Schiffler, Quivers with relations
arising from clusters (An case), Trans. Amer. Math. Soc. 358 (2006), no. 5,
1347–1364.
[29] Philippe Caldero and Bernhard Keller, From triangulated categories to cluster
algebras. II, Ann. Sci. École Norm. Sup. (4) 39 (2006), no. 6, 983–1009.
[30] Philippe Caldero and Bernhard Keller, From triangulated categories to cluster
algebras, Inv. Math. 172 (2008), 169–211.
[31] Philippe Caldero and Markus Reineke, On the quiver Grassmannian in the acyclic
case, Journal of Pure and Applied Algebra. 212 (2008), no. 11, 2369–2380.
[32] Henri Cartan and Samuel Eilenberg, Homological algebra, Princeton University
Press, Princeton, N. J., 1956.
[33] Frédéric Chapoton, Enumerative properties of generalized associahedra, Sém.
Lothar. Combin. 51 (2004/05), Art. B51b, 16 pp. (electronic).
[34] Frédéric Chapoton, Sergey Fomin, and Andrei Zelevinsky, Polytopal realizations
of generalized associahedra, Canad. Math. Bull. 45 (2002), no. 4, 537–566,
Dedicated to Robert V. Moody.
[35] Joe Chuang and Raphaël Rouquier, book in preparation.
156 Bernhard Keller
[57] Changjian Fu and Bernhard Keller, On cluster algebras with coefficients and
2-Calabi-Yau categories, Trans. Amer. Math. Soc. 362 (2010), 859–895.
[58] P. Gabriel, Unzerlegbare Darstellungen I, Manuscripta Math. 6 (1972), 71–103.
[59] P. Gabriel and A.V. Roiter, Representations of finite-dimensional algebras, Ency-
clopaedia Math. Sci., vol. 73, Springer–Verlag, 1992.
[60] Peter Gabriel, Auslander-Reiten sequences and representation-finite algebras,
Representation theory, I (Proc. Workshop, Carleton Univ., Ottawa, Ont., 1979),
Springer, Berlin, 1980, pp. 1–71.
[61] Christof Geiß, Bernard Leclerc, and Jan Schröer, Cluster algebra structures and
semicanonical bases for unipotent groups, arXiv:math/0703039.
[62] , Partial flag varieties and preprojective algebras, Ann. Jnst. Fourier.
(Grenoble) 58 (2008), no. 3, 825–876.
[63] , Preprojective algebras and cluster algebras, Trends in representation
theory and related topics, EMS Ser. Congr. Rep., Eur. Math. Soc., Zürich 2008,
pp. 253–283.
[64] , Semicanonical bases and preprojective algebras, Ann. Sci. École Norm.
Sup. (4) 38 (2005), no. 2, 193–253.
[65] , Rigid modules over preprojective algebras, Invent. Math. 165 (2006),
no. 3, 589–632.
[66] , Auslander algebras and initial seeds for cluster algebras, J. Lond. Math.
Soc. (2) 75 (2007), no. 3, 718–740.
[67] , Semicanonical bases and preprojective algebras. II. A multiplication
formula, Compos. Math. 143 (2007), no. 5, 1313–1334.
[68] Michael Gekhtman, Michael Shapiro, and Alek Vainshtein, On the properties
of the exchange graph of a cluster algebra, Math. Res. Lett. 15 (2008), no. 2,
321–330.
[69] , Cluster algebras and Poisson geometry, Mosc. Math. J. 3 (2003), no. 3,
899–934, 1199, {Dedicated to Vladimir Igorevich Arnold on the occasion of his
65th birthday}.
[70] , Cluster algebras and Weil-Petersson forms, Duke Math. J. 127 (2005),
no. 2, 291–311.
[71] Victor Ginzburg, Calabi-Yau algebras, arXiv:math/0612139v3 [math.AG].
[72] F. Gliozzi and R. Tateo, Thermodynamic Bethe ansatz and three-fold triangula-
tions, Internat. J. Modern Phys. A 11 (1996), no. 22, 4051–4064.
[73] Dieter Happel, On the derived category of a finite-dimensional algebra, Com-
ment. Math. Helv. 62 (1987), no. 3, 339–389.
[74] , Triangulated categories in the representation theory of finite-
dimensional algebras, Cambridge University Press, Cambridge, 1988.
[75] Dieter Happel, Idun Reiten, and Sverre O. Smalø, Tilting in abelian categories
and quasitilted algebras, Mem. Amer. Math. Soc. 120 (1996), no. 575, viii+ 88.
[76] Dieter Happel and Luise Unger, On a partial order of tilting modules, Algebr.
Represent. Theory 8 (2005), no. 2, 147–156.
[77] André Henriques, A periodicity theorem for the octahedron recurrence, J. Alge-
braic Combin. 26 (2007), no. 1, 1–26.
[78] David Hernandez and Bernard Leclerc, Cluster algebras and quantum affine
algebras, arXiv:0903.1452v1 [math.Q4].
158 Bernhard Keller
[79] Andrew Hubery, Acyclic cluster algebras via Ringel-Hall algebras, Preprint
available at the author’s home page.
[80] Osamu Iyama and Idun Reiten, Fomin-Zelevinsky mutation and tilting modules
over Calabi-Yau algebras, Amer. J. Math. 130 (2008), no. 4, 1087–1149.
[81] Osamu Iyama and Yuji Yoshino, Mutations in triangulated categories and rigid
Cohen-Macaulay modules, Inv. Math. 172 (2008), 117–168.
[82] Hiroshige Kajiura, Noncommutative homotopy algebras associated with open
strings, Rev. Math. Phys. 19 (2007), no. 1, 1–99.
[83] Masaki Kashiwara, Bases cristallines, C. R. Acad. Sci. Paris Sér. I Math. 311
(1990), no. 6, 277–280.
[84] Masaki Kashiwara and Pierre Schapira, Sheaves on manifolds, Grundlehren der
Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sci-
ences], vol. 292, Springer-Verlag, Berlin, 1994, With a chapter in French by
Christian Houzel, Corrected reprint of the 1990 original.
[85] Bernhard Keller, Deformed CY-completions, arXiv:0908.3499 [math.RT].
[86] , The periodicity conjecture for pairs of Dynkin diagrams,
arXiv:1001:1531.
[87] , Quiver mutation in Java, Java applet available at the author’s home
page.
[88] , Triangulated Calabi-Yau categories, Trends in Representation Theory
of Algebras (A. Skowronski, ed.), Eur. Math. Soc., 2008, pp. 467–489.
[89] , Deriving DG categories, Ann. Sci. École Norm. Sup. (4) 27 (1994),
no. 1, 63–102.
[90] , On triangulated orbit categories, Doc. Math. 10 (2005), 551–581.
[91] Bernhard Keller and Idun Reiten, Acyclic Calabi-Yau categories are clus-
ter categories, preprint, 2006, with an appendix by Michel Van den Bergh,
arXiv:math.RT/0610594, to appear in Compos. Math.
[92] , Cluster-tilted algebras are Gorenstein and stably Calabi-Yau, Advances
in Mathematics 211 (2007), 123–151.
[93] Bernhard Keller and Michel Van den Bergh, On two examples of Iyama and
Yoshino, arXiv:0803.0720.
[94] Bernhard Keller and Dong Yang, Derived equivalences from mutations of quivers
with potential, arXiv:0906:0761.
[95] Maxim Kontsevich, Donaldson-Thomas invariants, Mathematische Arbeitsta-
gung June 22-28, 2007, MPI Bonn, www.mpim-bonn.mpg.de/preprints/.
[96] , Donaldson-Thomas invariants, stability conditions and cluster trans-
formations, Report on joint work with Y. Soibelman, talks at the IHES, October
11 and 12, 2007.
[97] Maxim Kontsevich and Yan Soibelman, Notes on A-infinity algebras, A-infinity
categories and non-commutative geometry. I, arXiv:math.RA/0606241.
[98] , Stability structures, Donaldson-Thomas invariants and cluster trans-
formations, arXiv:0811.2435.
[99] Christian Krattenthaler, The F -triangle of the generalised cluster complex, Topics
in discrete mathematics, Algorithms Combin., vol. 26, Springer, Berlin, 2006,
pp. 93–126.
[100] A. Kuniba and T. Nakanishi, Spectra in conformal field theories from the Rogers
dilogarithm, Modern Phys. Lett. A 7 (1992), no. 37, 3487–3494.
Cluster algebras, representations, triangulated categories 159
[101] Atsuo Kuniba, Tomoki Nakanishi, and Junji Suzuki, Functional relations in solv-
able lattice models. I. Functional relations and representation theory, Internat.
J. Modern Phys. A 9 (1994), no. 30, 5215–5266.
[102] G. Lusztig, Canonical bases arising from quantized enveloping algebras, J. Amer.
Math. Soc. 3 (1990), no. 2, 447–498.
[103] , Total positivity in reductive groups, Lie theory and geometry, Progr.
Math., vol. 123, Birkhäuser Boston, Boston, MA, 1994, pp. 531–568.
[104] , Semicanonical bases arising from enveloping algebras, Adv. Math. 151
(2000), no. 2, 129–139.
[105] George Lusztig, Canonical bases and Hall algebras, Representation theories and
algebraic geometry (Montreal, PQ, 1997), NATO Adv. Sci. Inst. Ser. C Math.
Phys. Sci., vol. 514, Kluwer Acad. Publ., Dordrecht, 1998, pp. 365–399.
[106] Robert Marsh, Markus Reineke, and Andrei Zelevinsky, Generalized associa-
hedra via quiver representations, Trans. Amer. Math. Soc. 355 (2003), no. 10,
4171–4186 (electronic).
[107] Subir Mukhopadhyay and Koushik Ray, Seiberg duality as derived equivalence
for some quiver gauge theories, J. High Energy Phys. (2004), no. 2, 070, 22 pp.
(electronic).
[108] Gregg Musiker, A graph theoretic expansion formula for cluster algebras of
type Bn and Dn , arXiv:0710.3574v1 [math.CO]. to appear in the Annals of
Combinatorics.
[109] Hiraku Nakajima, Quiver varieties and cluster algebras, arXiv:0905002
[math.QA].
[110] L. A. Nazarova, Representations of quivers of infinite type, Izv. Akad. Nauk SSSR
Ser. Mat. 37 (1973), 752–791.
[111] Yann Palu, Grothendieck group and generalized mutation rule for 2-calabi-yau
triangulated categories, Journal of Pure and Applied Algebra. 213 (2009), no. 7,
1438–1449.
[112] , Cluster characters for 2-calabi-yau triangulated categories, Annales
de l’institut Fourier 58 (2008), no. 6, 2221–2248.
[113] F. Ravanini, A. Valleriani, and R. Tateo, Dynkin TBAs, Internat. J. Modern Phys.
A 8 (1993), no. 10, 1707–1727.
[114] Idun Reiten, Tilting theory and cluster algebras, preprint available at
www.institut.math.jussieu.fr/keller/ictp2006/lecturenotes/reiten.pdf.
[115] Claus Michael Ringel, The preprojective algebra of a quiver, Algebras and
modules, II (Geiranger, 1996), CMS Conf. Proc., vol. 24, Amer. Math. Soc.,
Providence, RI, 1998, pp. 467–480.
[116] , Some remarks concerning tilting modules and tilted algebras. Origin.
Relevance. Future., Handbook of Tilting Theory, LMS Lecture Note Series, vol.
332, Cambridge Univ. Press, Cambridge, 2007, pp. 49–104.
[117] C.M. Ringel, Tame algebras and integral quadratic forms, Lecture Notes in
Mathematics, vol. 1099, Springer Verlag, 1984.
[118] Kyoji Saito, Extended affine root systems. I. Coxeter transformations, Publ. Res.
Inst. Math. Sci. 21 (1985), no. 1, 75–179.
[119] Olivier Schiffmann, Lectures on Hall algebras, math.RT/0611617.
[120] Joshua S. Scott, Grassmannians and cluster algebras, Proc. London Math. Soc.
(3) 92 (2006), no. 2, 345–380.
160 Bernhard Keller
[121] Catherine Soanes and Angus Stevenson, Concise Oxford English Dictionary,
Oxford University Press, Oxford, 2004.
[122] James Dillon Stasheff, Homotopy associativity of H -spaces. I, II, Trans. Amer.
Math. Soc. 108 (1963), 275–292; ibid. 108 (1963), 293–312.
[123] Balázs Szendröi, Non-commutative Donaldson-Thomas theory and the conifold,
Geom. Toplo. 12 (2008), no. 2, 1171–1202.
[124] András Szenes, Periodicity of Y -systems and flat connections, Lett. Math. Phys.
89 (2009), no. 3, 217–230.
[125] Jean-Louis Verdier, Des catégories dérivées des catégories abéliennes,
Astérisque, vol. 239, Société Mathématique de France, 1996 (French).
[126] Jorge Vitória, Mutations vs. Seiberg duality, J. Algebra 321 (2009), no. 3, 816–
828.
[127] Alexandre Yu. Volkov, On the periodicity conjecture for Y -systems, Comm. Math.
Phys. 276 (2007), no. 2, 509–517.
[128] Keiichi Watanabe, Certain invariant subrings are Gorenstein. I, II, Osaka J.
Math. 11 (1974), 1–8; ibid. 11 (1974), 379–388.
[129] Shih-Wei Yang and Andrei Zelevinsky, Cluster algebras of finite type via Coxeter
elements and principal minors, Transform. Groups 13 (2008), no. 3–4, 855–895.
[130] Yuji Yoshino, Cohen-Macaulay modules over Cohen-Macaulay rings, London
Mathematical Society Lecture Note Series, vol. 146, Cambridge University Press,
Cambridge, 1990.
[131] Al. B. Zamolodchikov, On the thermodynamic Bethe ansatz equations for reflec-
tionless ADE scattering theories, Phys. Lett. B 253 (1991), no. 3–4, 391–394.
[132] Andrei Zelevinsky, Cluster algebras: notes for 2004 IMCC (Chonju, Korea,
August 2004), arXiv:math.RT/0407414.
[133] Andrei Zelevinsky, Mutations for quivers with potentials: Oberwolfach talk, april
2007, arXiv:0706.0822.
[134] , From Littlewood-Richardson coefficients to cluster algebras in three
lectures, Symmetric functions 2001: surveys of developments and perspectives,
NATO Sci. Ser. II Math. Phys. Chem., vol. 74, Kluwer Acad. Publ., Dordrecht,
2002, pp. 253–273.
[135] , Cluster algebras: origins, results and conjectures, Advances in algebra
towards millennium problems, SAS Int. Publ., Delhi, 2005, pp. 85–105.
[136] , What is a cluster algebra?, Notices of the A.M.S. 54 (2007), no. 11,
1494–1495.
Contents
1. Introduction 161
2. Categories of fractions and localization functors 164
3. Calculus of fractions 172
4. Localization for triangulated categories 178
5. Localization via Brown representability 192
6. Well generated triangulated categories 203
7. Localization for well generated categories 213
8. Epilogue: Beyond well generatedness 224
Appendix A. The abelianization of a triangulated category 225
Appendix B. Locally presentable abelian categories 227
References 233
1. Introduction
These notes provide an introduction to the theory of localization for triangulated
categories. Localization is a machinery to formally invert morphisms in a
category. We explain this formalism in some detail and we show how it is
applied to triangulated categories.
There are basically two ways to approach the localization theory for trian-
gulated categories and both are closely related to each other. To explain this,
let us fix a triangulated category T . The first approach is Verdier localization.
For this one chooses a full triangulated subcategory S of T and constructs
a universal exact functor T → T /S which annihilates the objects belonging
to S. In fact, the quotient category T /S is obtained by formally inverting all
morphisms σ in T such that the cone of σ belongs to S.
On the other hand, there is Bousfield localization. In this case one considers
an exact functor L : T → T together with a natural morphism ηX : X → LX
for all X in T such that L(ηX) = η(LX) is invertible. There are two full
triangulated subcategories arising from such a localization functor L. We have
161
162 Henning Krause
the subcategory Ker L formed by all L-acyclic objects, and we have the essential
image Im L which coincides with the subcategory formed by all L-local objects.
Note that L, Ker L, and Im L determine each other. Moreover, L induces an
∼
equivalence T / Ker L − → Im L. Thus a Bousfield localization functor T → T
is nothing but the composite of a Verdier quotient functor T → T /S with a
fully faithful right adjoint T /S → T .
Having introduced these basic objects, there are a number of immediate
questions. For example, given a triangulated subcategory S of T , can we find
a localization functor L : T → T satisfying Ker L = S or Im L = S? On the
other hand, if we start with L, which properties of Ker L and Im L are inherited
from T ? It turns out that well generated triangulated categories in the sense of
Neeman [33] provide an excellent setting for studying these questions.
Let us discuss briefly the relevance of well generated categories. The concept
generalizes that of a compactly generated triangulated category. For example,
the derived category of unbounded chain complexes of modules over some
fixed ring is compactly generated. Also, the stable homotopy category of CW-
spectra is compactly generated. Given any localization functor L on a compactly
generated triangulated category, it is rare that Ker L or Im L are compactly
generated. However, in all known examples Ker L and Im L are well generated.
The following theorem provides a conceptual explanation; it combines several
results from Section 7.
Theorem. Let T be a well generated triangulated category and S a full
triangulated subcategory which is closed under small coproducts. Then the
following are equivalent.
(1) The triangulated category S is well generated.
(2) The triangulated category T /S is well generated.
(3) There exists a cohomological functor H : T → A into a locally presentable
abelian category such that H preserves small coproducts and S = Ker H .
(4) There exists a small set S0 of objects in S such that S admits no proper full
triangulated subcategory closed under small coproducts and containing
S0 .
Moreover, in this case there exists a localization functor L : T → T such that
Ker L = S.
Note that every abelian Grothendieck category is locally presentable; in
particular every module category is locally presentable.
Our approach for studying localization functors on well generated trian-
gulated categories is based on the interplay between triangulated and abelian
structure. A well known construction due to Freyd provides for any triangulated
Localization theory for triangulated categories 163
indexed by all regular cardinals, such that for each α the category Aα (T )
is abelian and locally α-presentable in the sense of Gabriel and Ulmer [17].
Moreover, each inclusion Aα (T ) → A(T ) admits an exact right adjoint and the
composite
Hα : T −→ A(T ) −→ Aα (T )
Acknowledgement
The plan to write an introduction to the theory of triangulated localization
took shape during the “Workshop on Triangulated Categories” in Leeds 2006.
I wish to thank the organizers Thorsten Holm, Peter Jørgensen, and Raphaël
Rouquier for their skill and diligence in organizing this meeting. Most of these
notes were then written during a three months stay in 2007 at the Centre
de Recerca Matemàtica in Barcelona as a participant of the special program
“Homotopy Theory and Higher Categories”. I am grateful to the organizers
Carles Casacuberta, Joachim Kock, and Amnon Neeman for creating a stim-
ulating atmosphere and for several helpful discussions. Finally, I would like
164 Henning Krause
to thank Xiao-Wu Chen, Daniel Murfet, and Jan Šťovı́ček for their helpful
comments on a preliminary version of these notes.
Note that C[ −1 ] and Q are essentially unique if they exists. Now let us
sketch the construction of C[ −1 ] and Q . At this stage, we ignore set-theoretic
issues, that is, the morphisms between two objects of C[ −1 ] need not to form
a small set. We put Ob C[ −1 ] = Ob C. To define the morphisms of C[ −1 ],
consider the quiver (i.e. oriented graph) with set of vertices Ob C and with
set of arrows the disjoint union (Mor C) −1 , where −1 = {σ −1 : Y → X |
σ : X → Y }. Let P be the set of paths in this quiver (i.e. finite sequences
of composable arrows), together with the obvious composition which is the
concatenation operation and denoted by ◦P . We define Mor C[ −1 ] as the
quotient of P modulo the following relations:
θ : F ◦ G → Id D and η : Id C → G ◦ F
Proof. (1) ⇒ (2): Let D denote the full subcategory of C formed by all objects
X such that ηX is invertible. For each X ∈ D, let θ X : LX → X be the inverse
of ηX. Define F : C → D by F X = LX and let G : D → C be the inclusion.
We claim that F and G form an adjoint pair. In fact, it is straightforward to
check that the maps
D(F X, Y ) −→ C(X, GY ), α
→ Gα ◦ ηX,
and
C(X, GY ) −→ D(F X, Y ), β
→ θ Y ◦ Fβ,
are identity morphisms; see [27, IV.1]. We know from Proposition 2.3.1 that
θ is invertible because G is fully faithful. Therefore Lη = GF η is invertible.
Moreover, we have
C[ −1 ] WWWW
Q ggggg3
gg WWL̄WWW
ggggg WWW+
C WWWWWWW ∼
g ggggg3 C
WWWWW ggggg
F W W+ g gg G
D
where denotes the set of morphisms σ in C such that Lσ is invertible.
Proof. The characterization of a localization functor follows from Proposi-
tion 2.4.1. Now observe that equals the set of morphisms σ in C such that
F σ is invertible since G is fully faithful. Thus we can apply Proposition 2.3.1
to obtain the equivalence C[ −1 ] → D making the diagram commutative.
Thus ηX is invertible.
F G
(3) ⇔ (4): We use the factorization C −
→D− → C of L from Proposition 2.4.1.
Then we obtain for each W in C a factorization
∼ ∼
C(W, X) −→ C(W, LX) −→ C(F W, F X) −→ C(LW, LX)
of the map fW : C(W, X) → C(LW, LX) induced by L. Here, the first map is
induced by ηX, the second follows from the adjunction, and the third is induced
by G. Thus fW is bijective for all W iff the first map is bijective for all W iff
ηX is invertible.
(3) ⇒ (5): Take X = X.
F G
(5) ⇒ (1): We use again the factorization C − →D− → C of L from Propo-
sition 2.4.1. Fix σ in (L) and observe that F σ is invertible. Then we have
C(σ, X) ∼= C(σ, G(F X )) ∼ = D(F σ, F X ) and this implies that C(σ, X) is bijec-
tive since F σ is invertible.
(2) The morphism ηX belongs to (L) and factors uniquely through every
morphism X → Y in (L).
Proof. (1) ⇒ (2): Let (Xi )i∈I be a family of L-local objects. Thus the natural
morphisms Xi → LXi are invertible by Proposition 2.5.2 and they induce an
isomorphism
& ∼ & ∼ &
Xi −→ LXi −→ L( Xi ).
i i i
%
It follows that i Xi is L-local.
(2) ⇔ (3): We can identify C[(L)−1 ] = Im L by Corollary 2.5.3 and then
the right adjoint of the quotient functor identifies with the inclusion Im L → C.
Thus the right adjoint preserves small coproducts if and only if the inclusion
Im L → C preserves small coproducts.
(3) ⇒ (1): Write L as composite C − → C[(L)−1 ] − → C of the quotient
functor Q with its right adjoint L̄. Then Q preserves small coproducts since
it is a left adjoint. It follows that L preserves small coproducts if L̄ preserves
small coproducts.
and the essential image Im Γ equals the full subcategory of C consisting of all
Γ -colocal objects.
Remark 2.8.1. We think of Γ as L turned upside down; this explains our nota-
tion. Another reason for the use of Γ is the interpretation of local cohomology
as colocalization.
2.11. Notes
The category of fractions is introduced by Gabriel and Zisman in [18], but the
idea of formally inverting elements can be traced back much further; see for
instance [36]. The appropriate context for localization functors is the theory of
monads; see [27].
3. Calculus of fractions
3.1. Calculus of fractions
Let C be a category and a set of morphisms in C. The category of fractions
C[ −1 ] admits an elementary description if some extra assumptions on are
satisfied. We say that admits a calculus of left fractions if the following
holds.
X
α /Y
σ σ
X
α / Y
such that σ is in .
(LF3) Let α, β : X → Y be morphisms in C. If there is a morphism σ : X → X
in with α ◦ σ = β ◦ σ , then there exists a morphism τ : Y → Y in
with τ ◦ α = τ ◦ β.
Now assume that admits a calculus of left fractions. Then one obtains a
new category −1 C as follows. The objects are those of C. Given objects X
and Y , we call a pair (α, σ ) of morphisms
X
α / Y o σ
Y
Localization theory for triangulated categories 173
? Y1 _??
?? σ1
α1
??
??
α3
X? / Y3 o σ3
Y
?? O
??
?
α2 ?? σ
2
Y2
with σ3 in . The composition of two equivalence classes [α, σ ] and [β, τ ] is
by definition the equivalene class [β ◦ α, σ ◦ τ ] where σ and β are obtained
from condition (LF2) as in the following commutative diagram.
Z
β }}
}> `BB
BB σ
} BB
}} BB
}}
> Y `BB Z `
~ BB σ β ||
|> @ @@
α ~~ BB @@τ
~~ || @@
~ BB ||
~ |
X Y Z
We obtain a canonical functor
P : C −→ −1 C
by taking the identity map on objects and by sending a morphism α : X → Y
to the equivalence class [α, id Y ]. Let us compare P with the quotient functor
Q : C → C[ −1 ].
Proposition 3.1.1. The functor F : −1 C → C[ −1 ] which is the identity
map on objects and which takes a morphism [α, σ ] to (Q σ )−1 ◦ Q α is an
isomorphism.
Proof. The functor P inverts all morphisms in and factors therefore through
Q via a functor G : C[ −1 ] → −1 C. It is straightforward to check that
F ◦ G = Id and G ◦ F = Id.
From now on, we will identify −1 C with C[ −1 ] whenever admits a
calculus of left fractions. A set of morphisms in C admits a calculus of right
fractions if the dual conditions of (LF1) – (LF3) are satisfied. Moreover,
is called a multiplicative system if it admits both, a calculus of left fractions
and a calculus of right fractions. Note that all results about sets of morphisms
174 Henning Krause
admitting a calculus of left fractions have a dual version for sets of morphisms
admitting a calculus of right fractions.
Proof. We need to check the conditions (LF1) – (LF3). Observe first that
L = G ◦ F is a localization functor so that we can apply Proposition 2.5.2.
(LF1): This condition is clear because F is a functor.
σ α
(LF2): Let X ←−X− → Y be a pair of morphisms with σ in . This can be
completed to a commutative square
X
α /Y
σ σ
α
X / Y
(1) The functor Q has a right adjoint (which is then fully faithful).
(2) For each object X in C, there exist a morphism ηX : X → X such that X
is local with respect to and Q(ηX) is invertible.
α σ
Proof. The condition on Y implies that every fraction X → Y ← Y is equiv-
α σ
alent to one of the form X → Y ← Y with Y in S. Clearly, the fractions of
the form (α , σ ) with σ ∈ C(Y, Y ) and Y ∈ S form a small set.
is bijective. The map sends the equivalence class of a fraction to the equivalence
class of the same fraction. If [α, σ ] belongs to C[ −1 ](X, Y ) and τ is a mor-
phism with τ ◦ σ in ∩ D, then [τ ◦ α, τ ◦ σ ] belongs to D[( ∩ D)−1 ](X, Y )
and f sends it to [α, σ ]. Thus f is surjective. A similar argument shows that f
is injective.
is fully faithful.
Proof. Let (Xi )i∈I be a family of objects in C[ −1 ] which is indexed by a small
%
set I . We claim that the coproduct i Xi in C is also a coproduct in C[ −1 ].
Thus we need to show that for every object Y , the canonical map
&
C[ −1 ]( Xi , Y ) −→ C[ −1 ](Xi , Y ) (3.5.1)
i i
is bijective.
αi σi
To check surjectivity of (3.5.1), let (Xi → Zi ← Y )i∈I be a family of left
fractions. Using (LF2), we obtain a commutative diagram
% %
% αi % σi %
i
/ o i
i Xi i Zi i Y
πY
Zo
σ
Y
%
where πY : i Y → Y is the summation morphism and σ ∈ . It is easily
checked that
σ αi σi
(Xi → Z ← Y ) ∼ (Xi → Zi ← Y )
% σ αi
for all i ∈ I , and therefore (3.5.1) sends i Xi → Z ← Y to the family (Xi →
σi
Zi ← Yi )i∈I .
% α σ % α σ
To check injectivity of (3.5.1), let i Xi → Z ← Y and i Xi → Z ← Y
be left fraction such that
αi σ αi σ
(Xi → Z ← Y ) ∼ (Xi → Z ← Y )
commutative diagram
% %
Xi / Z
πZ
/Z
i i
%
i βi τ
%
Zi / Z∗
i
with τ ∈ .
¯ Thus τ ◦ σ ∈ ,¯ and we have
& α σ & α σ
( Xi → Z ← Y ) ∼ ( Xi → Z ← Y )
i i
% %
since πZ ◦ i αi = α and πZ ◦ i αi = α . Therefore the map (3.5.1) is also
injective, and this completes the proof.
3.6. Notes
The calculus of fractions for categories has been developed by Gabriel and
Zisman in [18] as a tool for homotopy theory.
X
α /Y β
/Z γ
/ SX
φ1 φ2 φ3 Sφ1
X
α / Y β
/ Z γ
/ SX
Localization theory for triangulated categories 179
X
α /Y β
/Z γ
/ SX
φ1 φ2 φ3 Sφ1
α
β γ
X / Y / Z / SX
of triangles.
(TR4) Given exact triangles (α1 , α2 , α3 ), (β1 , β2 , β3 ), and (γ1 , γ2 , γ3 ) with
γ1 = β1 ◦ α1 , there exists an exact triangle (δ1 , δ2 , δ3 ) making the fol-
lowing diagram commutative.
X
α1
/Y α2
/U α3
/ SX
β1 δ1
X
γ1
/Z γ2
/V γ3
/ SX
β2 δ2 Sα1
W W
β3
/ SY
β3 δ3
SY
Sα2
/ SU
is exact in U.
We have the following useful lemma.
Proof. We need to verify that admits a calculus of left and right fractions.
In fact, it is sufficient to check conditions (LF1) – (LF3), because then the dual
conditions are satisfied as well since the definition of is self-dual.
(LF1): This condition is clear because H is a functor.
(LF2): Let α : X → Y and σ : X → X be morphisms with σ in . We com-
plete α to an exact triangle and apply (TR3) to obtain the following morphism
between exact triangles.
W /X α /Y / SW
σ σ
W / X α / Y / SW
Proof. The proof is similar to that of Lemma 4.4.2; see [48, II.2.1.8] for
details.
T /S := T [(S)−1 ]
S
inc /T can / T /S
inc inc J
S
inc /T can / T /S
and ask when the functor J is fully faithful. We have the following criterion.
Proof. Suppose that condition (1) holds. We apply the criterion from
Lemma 3.4.1. Thus we take a morphism σ : Y → Y from (S) with Y in
T and need to find τ : Y → Y such that τ ◦ σ belongs to (S) ∩ T . To this
φ σ
end complete σ to an exact triangle X − → Y → SX. Then X belongs
→Y −
φ φ
to S and by our assumption we have a factorization X −
→Z−
→ Y of φ with
φ ψ
Z in S . Complete φ to an exact triangle Z −
→Y − → Y → SZ. Then (TR3)
yields a morphism τ : Y → Y satisfying ψ = τ ◦ σ . In particular, τ ◦ σ lies
in (S) ∩ T since Z belongs to S . The proof using condition (2) is dual.
and call them orthogonal subcategories with respect to S. Note that S ⊥ and
⊥
S are thick subcategories of T .
(1) Y belongs to S ⊥ .
(2) Y is (S)-local, that is, T (σ, Y ) is bijective for all σ in (S).
(3) The quotient functor induces a bijection T (X, Y ) → T /S(X, Y ) for all X
in T .
Proof. (1) ⇒ (2): Suppose T (X, Y ) = 0 for all X in S. Then every σ in (S)
induces a bijection T (σ, Y ) because T (−, Y ) is cohomological. Thus Y is
(S)-local.
(2) ⇒ (1): Suppose that Y is (S)-local. If X belongs to S, then the mor-
phism σ : X → 0 belongs to (S) and induces therefore a bijection C(σ, Y ).
Thus Y belongs to S ⊥ .
(2) ⇔ (3): Apply Lemma 3.2.2.
Localization theory for triangulated categories 185
where the first bijection follows from Lemma 4.8.1 and the others are clear
from the choice of F . Thus F ◦ Q is a left adjoint for the inclusion J .
It remains to show that ⊥ (S ⊥ ) = S. The inclusion ⊥ (S ⊥ ) ⊇ S is clear. Now
let X be an object of ⊥ (S ⊥ ). Then we have
T /S(QX, QX) ∼
= S ⊥ (F QX, F QX) ∼
= T (X, J (F QX)) = 0.
The following diagram displays the functors which arise from a localization
functor L : T → T . We use the convention that Fρ denotes a right adjoint of a
functor F .
Iρ Qρ
o o
S / T / T /S (L = Qρ ◦ Q and Γ = I ◦ Iρ )
I =inc Q=can
Localization theory for triangulated categories 187
T (X, Y ) ∼
= T (X, GF Y ) ∼
= T (F X, F Y ) = 0.
Now suppose that X is an object with T (X, Y ) = 0 for all L-local Y . Then
T (F X, F X) ∼
= T (X, GF X) = 0
Proof. (1) is a reformulation of Corollary 2.5.3, and (2) follows from Corol-
lary 2.4.2. (3) is an immediate consequence of the construction of Γ via Propo-
sition 4.9.1.
Proof. It suffices to prove (1) because (2) is the dual statement. So let L : T →
T be an exact localization functor. It follows from the construction of Γ that it
is of the form Γ = I ◦ Iρ where Iρ denotes a right adjoint of the fully faithful
inclusion I : Ker L → T . Thus Γ is a colocalization functor by Corollary 2.4.2.
The exactness of Γ follows from Lemma 4.2.1, and the identities Ker Γ = Im L
and Im Γ = Ker L are easily derived from the exact triangle (4.11.1).
4.13. Recollements
A recollement is by definition a diagram of exact functors
Iρ Qρ
o o
T o / T / T (4.13.1)
I o Q
Iλ Qλ
Proof. We apply Proposition 4.9.1 and its dual assertion. Observe first that any
localization functor L : T → T induces the following diagram.
Iρ =Γ Qρ =inc
o o
Ker L / T / Im L
I =inc Q=L
The functor I admits a left adjoint if and only if Q admits a left adjoint. Thus
the diagram can be completed to a recollement if and only if the inclusion I
admits a left adjoint.
Suppose now there is given a recollement of the form (4.13.1). Then L =
Qρ Q is a localization functor and the inclusion Ker L → T admits a left
∼
adjoint. The functor I induces an equivalence T − → Ker L and Qρ induces
∼
an equivalence T −→ Im L. It is straightforward to formulate and check the
various compatibilities of these equivalences.
As a final remark, let us mention that for any recollement of the form (4.13.1),
the functors Qλ and Qρ provide two (in general different) embeddings of T
into T . If we identify T = Im I , then Qρ identifies T with (T )⊥ and Qλ
identifies T with ⊥ (T ); see Proposition 4.10.1.
It follows that for each pair of chain complexes X, Y the set of mor-
phisms D(Mod A)(X, Y ) is small, since Qλ induces a bijection with
Localization theory for triangulated categories 191
where
( )
(z1 , z2 ) ∈ Eu and x ∈ U . Then we have short exact sequences 0 →
1
0 [0 1]
Z −−→ Eu −−−→ Z → 0 which yield pairwise different elements of Ext1A (Z, Z)
as u runs though the elements in U .
o o HomeAe (Ae,−)
Mod A/AeA o p∗ / Mod A / Mod eAe
o F
−⊗eAe eA
192 Henning Krause
o o RHomeAe (Ae,−)
Ker D(F ) o / D(A) / D(eAe)
inc
o D(F )
−⊗LeAe eA
4.17. Notes
Triangulated categories were introduced independently in algebraic geome-
try by Verdier in his thèse [48], and in algebraic topology by Puppe [38].
Grothendieck and his school used the formalism of triangulated and derived
categories for studying homological properties of abelian categories. Early
examples are Grothendieck duality and local cohomology for categories of
sheaves. The basic example of a triangulated category from topology is the
stable homotopy category.
Localizations of triangulated categories are discussed in Verdier’s thèse
[48]. In particular, he introduced the localization (or Verdier quotient) of a
triangulated category with respect to a triangulated subcategory. In the context
of stable homotopy theory, it is more common to think of localization functors
as endofunctors; see for instance the work of Bousfield [8], which explains
the term Bousfield localization. The standard reference for recollements is [6].
Resolutions of unbounded complexes were first studied by Spaltenstein in [44];
see also [5].
is surjective.
Proof. For a proof of (1) see [24, Theorem A]. To prove (2), suppose that F
preserves small coproducts. Then one defines the right adjoint G : U → T by
sending an object X in U to the object in T representing U(F −, X). Thus
U(F −, X) ∼= T (−, GX). Conversely, given a right adjoint of F , it is automatic
that F preserves small coproducts.
(1) The functor F admits a right adjoint if and only if F preserves small
coproducts.
(2) The functor F admits a left adjoint if and only if F preserves small products.
Localization theory for triangulated categories 195
where the last isomorphism uses that F X is compact. It is easily checked that
the objects X in T such that T (X , φ) is an isomorphism form a localizing
subcategory of T . Thus φ is an isomorphism because the compact objects
generate T .
(3) The right adjoint of the inclusion functor Ker L → T preserves small
coproducts.
(4) The right adjoint of the quotient functor T → T / Ker L preserves small
coproducts.
(5) The subcategory Im L of all L-local objects is closed under taking small
coproducts.
where I denotes the inclusion and Q a right adjoint of the inclusion Ker L → T .
∼
Note that Q induces an equivalence T / Im L − → Ker L; see Propositions 4.11.1
and 4.12.1. The functors I and Q have left adjoints. Thus the pair (I, Q) gives
rise to a recollement if and only if I and Q both admit right adjoints. It follows
from Proposition 4.9.1 that this happens if and only if Q admits a right adjoint.
Now Brown’s representability theorem implies that this is equivalent to the
fact that Q preserves small coproducts. And Proposition 3.5.1 shows that Q
preserves small coproducts if and only if Im L is closed under taking small
coproducts. This finishes the proof.
Remark 5.5.2. (1) The implication (6) ⇒ (5) holds without any extra assump-
tion on T .
(2) Suppose an exact localization functor L : T → T preserves small
coproducts. If T is compactly generated, then Im L is compactly generated. This
follows from Lemma 5.4.1, because the left adjoint of the inclusion Im L → T
sends the compact generators of T to compact generators for Im L. A similar
argument shows that Im L is perfectly generated provided that T is perfectly
generated.
Example 5.5.3. Let S be the stable homotopy category of spectra and ∧ its
smash product. Then an exact localization functor L : S → S preserves small
coproducts if and only if L is of the form L = − ∧ E for some spectrum E. We
Localization theory for triangulated categories 197
sketch the argument. Let S denote the sphere spectrum. There exists a natural
morphism ηX : X ∧ LS → LX for each X in S. Suppose that L preserves
small coproducts. Then the subcategory of objects X in S such that ηX is
invertible contains S and is closed under forming small coproduts and exact
triangles. Thus L = − ∧ E for E = LS.
Sc
inc /Tc can / T c /S c
inc inc J
I =inc
S /T Q=can
/ T /S
Proof. The inclusion I preserves compactness and therefore the right adjoint Iρ
preserves small coproducts by Lemma 5.4.1. Thus Qρ preserve small coprod-
ucts by Proposition 5.5.1, and therefore Q preserves compactness, again by
Lemma 5.4.1. It follows that J induces a functor T c /S c → (T /S)c . In partic-
ular, Q sends a set of compact generators of T to a set of compact generators
for T /S.
Next we apply Lemma 4.7.1 to show that J is fully faithful. For this, one
needs to check that every morphism from a compact object in T to an object
in S factors through some object in S c . This follows from Theorem 7.2.1.
The image of J is a full triangulated subcategory of T c which generates T /S.
Another application of Corollary 7.2.2 shows that every compact object of T /S
is a direct factor of some object in the image of J .
Let L : T → T denote the localization functor with Ker L = S. Then S ⊥
equals the full subcategory of L-local objects. This subcategory is closed under
small coproducts since S is generated by compact objects. Thus the existence
of the recollement follows from Proposition 5.5.1.
H ∗ : T −→ A
1 All graded rings and modules are graded over Z. Morphisms between graded modules are
degree zero maps.
Localization theory for triangulated categories 199
T
L̃ /T
H∗ H∗
A
L /A
Now consider the essential image Im L of L which equals the full subcate-
gory formed by all L-local objects in A. Because L is exact, this subcategory
is coherent, that is, for any exact sequence X1 → X2 → X3 → X4 → X5 with
X1 , X2 , X4 , X5 ∈ A, we have X3 ∈ A. This is an immediate consequence of
the 5-lemma. In addition, Im L is closed under taking small products. The L-
local objects form an abelian Grothendieck category and therefore Im L admits
an injective cogenerator, say I ; see [16]. Using again Brown’s representability
theorem, there exists I˜ in T such that
A(H ∗ −, I ) ∼
= T (−, I˜) and therefore A(H ∗ −, I )∗ ∼
= T (−, I˜)∗ . (5.7.1)
Now consider the subcategory V of T which is formed by all objects X in T
such that H ∗ X is L-local. This is a triangulated subcategory which is closed
under taking small products. Observe that I˜ belongs to V. To prove this, take a
free presentation
F1 −→ F0 −→ H ∗ C −→ 0
over and apply A(−, I )∗ to it. Using the isomorphism (5.7.1), we see that
H ∗ I˜ belongs to Im L because Im L is coherent and closed under taking small
products.
200 Henning Krause
We claim that LH ∗ η̃X and ηH ∗ L̃X are invertible for each X in T . The mor-
phism η̃X induces an exact triangle
η̃X
X → X −→ L̃X → SX
with L̃X = 0 = L̃SX . Applying the cohomological functor LH ∗ , we see that
LH ∗ η̃X is an isomorphism, since LH ∗ X = 0 = LH ∗ SX . Thus LH ∗ η̃ is
invertible. The morphism ηH ∗ L̃X is invertible because H ∗ L̃X is L-local. This
follows from the fact that L̃X belongs to U.
The commutative square (5.7.2) implies that H ∗ η̃X is invertible if and only
if ηH ∗ X is invertible. Thus if X is L̃-local, then H ∗ X is L-local. The converse
holds if H ∗ reflects isomorphisms.
Remark 5.7.2. (1) The localization functor L̃ is essentially uniquely deter-
mined by H ∗ and L, because Ker L̃ = Ker LH ∗ .
(2) Suppose that C is a generator of T . If L preserves small coproducts,
then it follows that L̃ preserves small coproducts. In fact, the assumption
Localization theory for triangulated categories 201
K(Mod A)(A, S n X) ∼
= H n X.
K(Mod A)(S n X, I ) ∼
= HomA (H n X, I ).
with the inclusion Coloc I → K(Mod A) is the right adjoint Qρ of Q and takes
a complex to its K-injective resolution.
o RHomA (B,−)
o
D(B) o / D(A) / U
f∗
o
−⊗LA B
The triangulated category U is equivalent to the kernel of − ⊗LA B, and one can
show that Ker(− ⊗LA B) is the localizing subcategory of D(A) generated by the
complexes of the form
x
··· → 0 → A −
→ A → 0 → ··· (x ∈ S).
5.10. Notes
The Brown representability theorem in homotopy theory is due to Brown
[9]. Generalizations of the Brown representability theorem for triangulated
categories can be found in work of Franke [13], Keller [21], and Neeman [31,
33]. The version used here is taken from [24]. The finite localization theorem
for compactly generated triangulated categories is due to Neeman [30]; it is
based on previous work of Bousfield, Ravenel, Thomason-Trobaugh, Yao, and
others. The cohomological localization functors commuting with localization
functors of graded modules have been used to set up local cohomology functors
in [7].
Localization theory for triangulated categories 203
is an isomorphism.
C5
γ
/ C
55
μ 5
5
5 μ
X
Analogously, one defines for a morphism φ : X → X in T the category S/φ
whose objects are commuting squares in T of the form
C
γ
/ C
μ μ
X
φ
/ X
with C, C ∈ S.
Proof. Straightforward.
X1
φ1
/ X2 φ2
/ X3
μ1 μ2 μ3
φ1 φ2
X1 / X / X
2 3
C1
γ1
/ C2 γ2
/ C3
id id δ
C1
γ1
/ C2 δγ2
/ C
3
μ1 μ2 μ3
X1
φ1
/ X2 φ2
/ X3
C1
γ1
/ C2 γ2
/ C3
( ) ( )
ρ id
id id ( ) 0
γ2
C̄1 C1
[ γ̄1 0 ]
/ C2 0
/ C3 SC1
[ σ μ1 −σρ ] μ2 [ μ3 0 ]
X1
φ1
/ X2 φ2
/ X3
T and denote by S/(φ1 , φ2 ) the category whose objects are the commutative
diagrams in T of the following form.
C1
γ1
/ C2 γ2
/ C3
X1
φ1
/ X2 φ2
/ X3
such each Ci belongs to S. Then the full subcategory formed by the diagrams
γ1 γ2 γ3
such that there exists an exact triangle C1 −
→ C2 −→ C3 − → SC1 is a cofinal
subcategory of S/(φ1 , φ2 ).
S
hS
/ Addα (S op , Ab)
f =inc f∗
T
hT
/ A(T )
208 Henning Krause
(1) f ∗ is fully faithful and identifies Addα (S op , Ab) with the full subcategory
formed by all colimits of objects in {T (−, X) | X ∈ S}.
(2) f∗ preserves small coproducts if and only if (PG2) holds for S.
(3) Suppose that S is a triangulated subcategory of T . Then for X in T , the
adjunction morphism f ∗ f∗ (hT X) → hT X identifies with the canonical
morphism
colim hT C −→ hT X.
−−−→
(C,μ)∈S/X
colim T (−, C)
−−−→
(C,μ)∈T α /X
Proof. It is clear that (1) and (2) are equivalent, and it follows from Proposi-
tion 6.7.1 that (2) implies (3). To prove that (3) implies (2), assume that H pre-
%
serves small coproducts. Let φ : X → i∈I Yi be a morphism in T with X ∈ S.
% %
Write i∈I Yi = colim YJ as α-filtered colimit of coproducts YJ = i∈J Yi
−−−→
J ⊆I
with card J < α. Then we have
colim T (X, YJ ) ∼
= colim HomS (S(−, X), H YJ )
−−−→ −−−→
J ⊆I J ⊆I
∼
= HomS (S(−, X), colim H YJ )
−−−→
J ⊆I
&
∼
= HomS (S(−, X), H Yi )
i∈I
&
∼
= HomS (S(−, X), H ( Yi ))
i∈I
&
∼
= T (X, Yi ).
i∈I
210 Henning Krause
Thus φ factors through some YJ , and it follows that X is α-small. Now Propo-
sition 6.7.1 implies that (PG2) holds for S.
Proof. The left hand term of (6.9.1) defines a functor H̃ : T → A and we need
to show that the canonical morphism H̃ → H is invertible.
First observe that H̃ is cohomological. This is a consequence of Proposi-
tion 6.6.2 and Lemma 6.4.1, because for any exact triangle X1 → X2 → X3 →
SX1 in T , the sequence H̃ X1 → H̃ X2 → H̃ X3 can be written as α-filtered
colimit of exact sequences in A.
Next we claim that H̃ preserves small coproducts. To this end consider the
exact functor H̄ : A(T ) → A which extends H ; see Lemma A.2. Note that H̄
preserve small coproducts because H has this property. We have for X in T
H̃ X = colim H̄ T (−, C) ∼ = H̄ colim T (−, C) .
−−−→α −−−→α
(C,μ)∈T /X (C,μ)∈T /X
Hα : T −→ Aα (T ), X
→ colim T (−, C),
−−−→
(C,μ)∈T α /X
hα : T −→ Addα ((T α )op , Ab), X
→ T (−, X)|T α ,
α op
2 Addα ((T ) , Ab)
eeeeeeeeeee
hα
T YeYYYYYYY ∼ f∗
YYYYYY
Hα YY,
Aα (T )
hA(T α )
Tα
hT α
/ A(T α ) / Aα (T )
Remark 6.10.2. The universal property can be used to show that the category
T α of α-compact objects does not depend on the choice of a perfectly generating
set for T . More precisely, if T is α-well generated, then two α-localizing
subcategories coincide if each contains a small set of α-small perfect generators.
This follows from the fact that the functor Hα identifies the α-compact objects
with the α-presentable projective objects of Aα (T ).
6.11. Notes
Well generated triangulated were introduced and studied by Neeman in his
book [33] as a natural generalization of compactly generated triangulated cat-
egories. For an alternative approach which simplifies the definition, see [23].
More recently, well generated categories with specific models have been stud-
ied; see [37, 47] for work involving algebraic models via differential graded
categories, and [20] for topological models. In [43], Rosický used combina-
torial models and showed that there exist universal cohomological functors
into locally presentable categories which are full. Interesting consequences of
this fact are discussed in [35]. The description of the universal cohomological
functors in terms of filtered colimits seems to be new.
Localization theory for triangulated categories 213
Ui
ιi
/ Xi πi
/X / SUi
σi σ Sσi
Vi / Xi /Y / SVi
On the other hand, the exact triangle (7.1.1) induces the exact sequence
& H (id −φi ) & Hψ &
H Xi −−−−−→ H Xi −→ H X −→ H S Xi ,
i∈N i∈N i∈N
Xi 6
φi
/ Xi+1
66
6
πi 6
6 i+1
π
Y
Localization theory for triangulated categories 215
Proof. The assumption Ui := Ker πi = Ker φi implies that there exists a mor-
phism πi : Y → Xi with πi πi = id Y and φi πi = πi+1
for all i ≥ 1. Thus we
have a sequence of commuting squares
[ inc πi ] / Xi
Ui Y
( )
0 0 φi
0 id
[ inc πi+1 ]
Ui+1 Y / Xi+1
where the horizontal maps are isomorphisms. Taking colimits on both sides,
the assertion follows.
Sα
inc /Tα can / T α /S α
inc inc J
S
inc /T can / T /S
216 Henning Krause
S
inc /T
H hα
∗
Addα (C op , Ab)
i / Addα ((T α )op , Ab)
H : S −→ Addα (C op , Ab), X
→ S(−, X)|C .
T
Q=can
/ T /S
hα K
q∗
Addα ((T α )op , Ab) / Addα ((T α /S α )op , Ab)
Proof. In the preceding proof of Theorem 7.2.1, we can choose for C instead
of S ∩ T α the α-localizing subcategory of T which is generated by S0 . Then
the proof shows that C provides a small set of α-small perfect generators for S.
Thus we have S α = C by definition.
φ
L : T → T inverting φ. To see this, complete φ to an exact triangle X − →
Y → Z → SX and let L be the localization functor such that Ker L equals
the localizing subcategory generated by Z. Conversely, any exact localization
functor L : T → T is the universal exact localization functor inverting some
morphism φ provided that Ker L is generated by a small set S0 of objects. To
%
see this, take φ : 0 → X∈S0 X.
(2) Let T be a triangulated category and L : T → T an exact localization
functor such that S = Ker L is generated by a single object W . Then the
first morphism Γ X → X from the functorial triangle Γ X → X → LX →
S(Γ X) is called cellularization and the second morphism X → LX is called
nullification with respect to W . The objects in S are built from W .
hβ (T ) hβ (U) hβ (T )
∗
Addβ ((T β )op , Ab)
f
/ Addβ ((U β )op , Ab) f∗
/ Addβ ((T β )op , Ab)
Observe that Theorem 7.4.1 provides a partial answer to the telescope con-
jecture for compactly generated categories. This conjecture claims that the
kernel of a localization functor L : T → T is generated by compact objects
provided that L preserves small coproducts. Part (1) implies that S = Ker L
is generated by morphisms between compact objects, and part (2) says that
S is generated by ℵ1 -compact objects. I am grateful to Amnon Neeman for
explaining to me how to deduce (2) from (1). The following corollary makes
the connection with the telescope conjecture more precise; just put α = ℵ0 .
∼ h̄∗ ∼
h∗ : Addβ ((T β )op , Ab) −
→ Lexβ (A(T β )op , Ab) −
→ Lexβ ((Aβ )op , Ab) −
→A
where the first equivalence follows from Lemma B.1 and the second equivalence
follows from Lemma B.6. The functor h̄∗ is a left Kan extension; it takes a
filtered colimit
Note that h∗ is exact and preserves small coproducts. This follows from
Lemma B.5 and the fact that h̄∗ is left adjoint to the restriction functor h̄∗ .
The composite h∗ ◦ hβ : T → A coincides with H on T β and therefore
h ∗ ◦ hβ ∼
= H by Theorem 6.9.1. In particular, we have for each X in T that
H X = 0 if and only if h∗ (hβ X) = 0. Now we use the same argument as in
the proof of Theorem 7.4.1 and show that Ker H is generated by all homotopy
colimits of countable sequences of morphisms in T β which are annihilated by
H.
T
L /T
Hα Hα
L
Aα (T ) / Aα (T )
∼
K(Mod A)/(Ker K(Q)) −→ K(A)
since K(Q) has K(T ) as a fully faithful right adjoint. Moreover, the cohomology
of each object in the kernel of K(Q) lies in the kernel of Q. Thus we obtain the
following commutative diagram.
Ker K(Q)
inc / K(Mod A) K(Q)
/ K(A)
can F
S
inc / D(A) can / D(A)/S
H∗ H̄
Ker Q
inc / Mod A Q
/A
It is easily checked that the kernel of F consists of all acyclic complexes. Thus
∼
F induces an equivalence D(A) − → D(A)/S.
7.8. Notes
Given a triangulated category T , there are two basic questions when one
studies exact localization functors T → T . One can ask for the existence of
a localization functor with some prescribed kernel, and one can ask for a
classification, or at least some structural results, for the set of all localization
functors on T . Well generated categories provide a suitable setting for some
partial answers.
The fact that cohomological functors induce localization functors is well
known for compactly generated triangulated categories [28], but the result
seems to be new for well generated categories. The localization theorem
which describes the localization with respect to a small set of objects is
due to Neeman [33]. The example of the derived category of an abelian
Grothendieck category is discussed in [3, 34]. The description of the ker-
nel of an exact functor between well generated categories seems to be new. A
motivation for this is the telescope conjecture which is due to Bousfield and
Ravenel [8, 40] and originally formulated for the stable homotopy category of
CW-spectra.
It is interesting to note that the existence of localization functors depends to
some extent on axioms from set theory; see for instance [11, 10].
224 Henning Krause
C(−, X) −→ C(−, Y ) −→ F −→ 0.
The morphisms between two coherent functors form a small set by Yoneda’s
lemma, and the coherent functors C op → Ab form an additive category with
cokernels. We denote this category by C.
A basic tool is the fully faithful Yoneda functor hC : C → C which sends
an object X to C(−, X). One might think of this functor as the completion of
C with respect to the formation of finite colimits. To formulate some further
properties, we recall that a morphism X → Y is a weak kernel for a morphism
Y → Z if the induced sequence C(−, X) → C(−, Y ) → C(−, Z) is exact.
to the cokernel of H φ.
(2) The category C has cokernels, and it is therefore sufficient to show
that C has kernels. To this end fix a morphism F1 → F2 with the following
presentation.
C(−, X1 ) / C(−, Y1 ) / F1 /0
C(−, X2 ) / C(−, Y2 ) / F2 /0
226 Henning Krause
C(−, X1 ) / C(−, Y1 ) / F1 /0
Notes
The abelianization of a triangulated category appears in Verdier’s thèse [48]
and in Freyd’s work on the stable homotopy category [15]. Note that their
construction is slightly different from the one given here, which is based on
coherent functors in the sense of Auslander [4].
Lemma B.1. Let C be a small additive category with α-coproducts. Then the
Yoneda functor induces an equivalence
∼
Lexα (Cop , Ab) −→ Addα (C op , Ab)
by taking a functor F to F ◦ hC .
colim C(−, C) −→ F
−−−→
(C,μ)∈C/F
(2) The functor F belongs to Lexα (C op , Ab) if and only if the category C/F is
α-filtered.
Proof. Combine Yoneda’s lemma with the fact that the inclusion
Lexα (C op , Ab) → Add(C op , Ab) preserves α-filtered colimits.
There is a general result for the category Lexα (C op , Ab) which says that tak-
ing α-filtered colimits commutes with taking α-limits; see [17, Korollar 7.12].
Here we need the following special case.
Lemma B.4. Suppose the category Lexα (C op , Ab) is abelian. Then an α-filtered
colimit of exact sequences is again exact.
Proof. We need to show that taking α-filtered colimits commutes with taking
kernels and cokernels. A cokernel is nothing but a colimit and therefore taking
colimits and cokernels commute. The statement about kernels follows from
the fact that the inclusion Lexα (C op , Ab) → Add(C op , Ab) preserves kernels
and α-filtered colimits. Thus we can compute kernels and α-filtered colimits
in Add(C op , Ab) and therefore in the category Ab of abelian groups. In Ab it is
well known that taking kernels and filtered colimits commute.
Lemma B.5. Suppose that C is abelian. Then Lexα (C op , Ab) is abelian and the
Yoneda functor hC : C → Lexα (C op , Ab) is exact. Given an abelian category
A which admits small coproducts and exact α-filtered colimits, and given a
functor F : Lexα (C op , Ab) → A preserving α-filtered colimits, we have that F
is exact if and only if F ◦ hC is exact.
Proof. We use the analogue of Lemma B.2 for morphisms which says that each
morphism φ in Lexα (C op , Ab) can be written as α-filtered colimit φ = colim φi
−−−→
i∈C/φ
230 Henning Krause
and we see that Lexα (C op , Ab) is abelian; see Lemma B.4. The formula for
kernels and cokernels shows that each exact sequence can be written as α-
filtered colimit of exact sequences in the image of the Yoneda embedding. The
criterion for the exactness of a functor Lexα (C op , Ab) → A is an immediate
consequence.
f∗ : A −→ Lexα (C op , Ab), X
→ A(f −, X),
Proof. The functor is the left Kan extension of f ; it takes F = colim C(−, C)
−−−→
(C,μ)∈C/F
in Lexα (C op , Ab) to colim f C in A. We refer to [17, Satz 7.8] for details.
−−−→
(C,μ)∈C/F
Proof. The implications (1) ⇒ (2) ⇒ (3) are clear. So we prove (3) ⇒ (1). It
is convenient to identify Addα (C op , Ab) with Lexα (Cop , Ab) and this identifies
F with the left exact functor F̄ : C → Ab which extends F . In fact, F̄ is exact
since F is cohomological, by Lemma A.2. Now write F̄ as α-filtered colimit of
representable functors F̄ = colim C(−, M); see Lemma B.2. The exactness
−−−→
F̄
(M,ν)∈C/
of F̄ implies that the representable functors C(−, C) with C ∈ C form a full
F̄ which is cofinal. We identify this subcategory with C/F
subcategory of C/
and conclude from Lemma 6.4.1 that C/F is α-filtered.
has a left adjoint f ∗ which sends C(−, X) to D(−, f X) for all X in C. Moreover,
the following holds.
Proof. The left adjoint of f∗ is the left Kan extension. We can describe it explic-
itly if we identify Addα (C op , Ab) with Lexα (Cop , Ab); see Lemma B.1. Given a
functor F in Lexα (Cop , Ab) written as α-filtered colimit F = colim C(−, C)
−−−→
(C,μ)∈C/F
of representable functors, we put
f ∗ F = colim D(−, fC).
−−−→
(C,μ)∈C/F
232 Henning Krause
hC =
C
hC
/ C / Lex (Cop , Ab) / Addα (C op , Ab)
α
f f f∗ f∗
hD
hD
/D / Lex (D = / Addα (Dop , Ab)
D α
op , Ab)
X), G) = HomD(D(−,
HomD(f ∗ C(−, fX), G) ∼
= G(fX)
= f∗ G(X) ∼ X), f∗ G)
= HomC(C(−,
for all G in Lexα (Dop , Ab). Clearly, this isomorphism extends to every colimit
of representable functors.
(1) The exactness of f∗ is clear because a sequence F → F → F in
Addα (C op , Ab) is exact if and only if F X → F X → F X is exact for all X in
C. For the exactness of f ∗ we identify again Add(C op , Ab) with Lexα (Cop , Ab)
and apply Lemma B.5. Thus we need to check that the composition of f ∗ with
the Yoneda functor hC is exact. But we have that f ∗ ◦ hC = hD ◦ f, and now the
exactness follows from that of f . Finally, we use the fact that taking α-filtered
colimits in Addα (Dop , Ab) is exact by Lemma B.4.
(2) It is well known that for any epimorphism f : C → D of addi-
tive categories inducing a bijection Ob C → Ob D, the restriction functor
Add(Dop , Ab) → Add(C op , Ab) is fully faithful; see [29, Corollary 5.2]. Given
a triangulated subcategory C ⊆ C, the quotient functor C → C/C is an epimor-
phism. Thus the assertion follows since Addα (C op , Ab) is a full subcategory of
Add(C op , Ab).
(3) We keep our identification Addα (C op , Ab) = Lexα (Cop , Ab) and consider
the adjunction morphism η : Id → f∗ ◦ f ∗ . We claim that η is an isomorphism.
Because f is fully faithful, ηF is an isomorphism for each representable functor
X). It follows that ηF is an isomorphism for all F since f ∗ and f∗
F = C(−,
both preserve α-filtered colimits and each F can be expressed as α-filtered
colimit of representable functors. Now Proposition 2.3.1 implies that f ∗ is
fully faithful.
Let F be a cohomological functor in Addα (Dop , Ab) and apply Lemma B.7 to
write the functor as α-filtered colimit F = colim D(−, D) of representable
−−−→
(D,μ)∈D/F
functors. Suppose first that every morphism D(−, D) → F factors through
D(−, f C) for some C ∈ C. Then Im f/F is a cofinal subcategory of D/F
Localization theory for triangulated categories 233
essential image is closed under taking colimits. Now suppose that F belongs
to the essential image of f ∗ . Then F = f ∗ G ∼
= f ∗ f∗ f ∗ G = f ∗ f∗ F for some
G. The functor f∗ F is cohomological and therefore f∗ F = colim C(−, C),
−−−→
(C,μ)∈C/f∗ F
∼
again by Lemma B.7. Thus F = colim D(−, f C) and we use Lemma B.3
−−−→
(C,μ)∈C/f∗ F
to conclude that each morphism D(−, D) → F factors through D(−, f C) for
some (C, μ) ∈ C/f∗ F .
(4) Let F be a cohomological functor in Addα (C op , Ab) and apply
Lemma B.7 to write the functor as α-filtered colimit F = colim C(−, C)
−−−→
(C,μ)∈C/F
of representable functors. Now f ∗ F = colim D(−, f C) = 0 if and only
−−−→
(C,μ)∈C/F
if for each D ∈ D, we have colim D(D, f C) = 0. This happens iff for
−−−→
(C,μ)∈C/F
each (C, μ) ∈ C/F , we find a morphism γ : C → C in C/F inducing a map
D(f C, f C) → D(f C, f C ) which annihilates the identity morphism. But this
means that f γ = 0 and that μ : C(−, C) → F factors through C(−, γ ).
Notes
Locally presentable categories were introduced and studied by Gabriel and
Ulmer in [17]; see [1] for a modern treatment. In [33], Neeman initiated the use
of locally presentable abelian categories for studying triangulated categories.
References
[1] J. Adámek and J. Rosický, Locally presentable and accessible categories,
Cambridge Univ. Press, Cambridge, 1994.
[2] J. F. Adams, Stable homotopy and generalised homology, Univ. Chicago Press,
Chicago, Ill., 1974.
[3] L. Alonso Tarrı́o, A. Jeremı́as López and M. J. Souto Salorio, Localization in
categories of complexes and unbounded resolutions, Canad. J. Math. 52 (2000),
no. 2, 225–247.
[4] M. Auslander, Coherent functors, in Proc. Conf. Categorical Algebra (La Jolla,
Calif., 1965), 189–231, Springer, New York, 1966.
[5] L. Avramov and S. Halperin, Through the looking glass: a dictionary between
rational homotopy theory and local algebra, in Algebra, algebraic topology and
their interactions (Stockholm, 1983), 1–27, Lecture Notes in Math., 1183, Springer,
Berlin, 1986.
234 Henning Krause
[30] A. Neeman, The connection between the K-theory localization theorem of Thoma-
son, Trobaugh and Yao and the smashing subcategories of Bousfield and Ravenel,
Ann. Sci. École Norm. Sup. (4) 25 (1992), no. 5, 547–566.
[31] A. Neeman, The Grothendieck duality theorem via Bousfield’s techniques and
Brown representability, J. Amer. Math. Soc. 9 (1996), no. 1, 205–236.
[32] A. Neeman, Non-compactly generated categories, Topology 37 (1998), no. 5,
981–987.
[33] A. Neeman, Triangulated categories, Ann. of Math. Stud., 148, Princeton Univ.
Press, Princeton, NJ, 2001.
[34] A. Neeman, On the derived category of sheaves on a manifold, Doc. Math. 6
(2001), 483–488 (electronic).
[35] A. Neeman, Brown representability follows from Rosický, preprint (2007).
[36] O. Ore, Linear equations in non-commutative fields, Ann. of Math. (2) 32 (1931),
no. 3, 463–477.
[37] M. Porta, The Popescu-Gabriel theorem for triangulated categories,
arXiv:0706.4458v1.
[38] D. Puppe, On the structure of stable homotopy theory, in Colloquium on algebraic
topology, Aarhus Universitet Matematisk Institut (1962), 65–71.
[39] D. G. Quillen, Homotopical algebra, Lecture Notes in Math., 43, Springer, Berlin,
1967.
[40] D. C. Ravenel, Localization with respect to certain periodic homology theories,
Amer. J. Math. 106 (1984), no. 2, 351–414.
[41] J. Rickard, Idempotent modules in the stable category, J. London Math. Soc. (2)
56 (1997), no. 1, 149–170.
[42] J. Rickard, Bousfield localization for representation theorists, in Infinite length
modules (Bielefeld, 1998), 273–283, Birkhäuser, Basel.
[43] J. Rosický, Generalized Brown representability in homotopy categories, Theory
Appl. Categ. 14 (2005), no. 19, 451–479 (electronic).
[44] N. Spaltenstein, Resolutions of unbounded complexes, Compositio Math. 65
(1988), no. 2, 121–154.
[45] B. Stenström, Rings of quotients, Springer, New York, 1975.
[46] J. Šťovı́ček, Locally well generated homotopy categories of complexes,
arXiv:0810.5684v1.
[47] G. Tabuada, Homotopy theory of well-generated algebraic triangulated categories,
J. K-theory, in press.
[48] J.-L. Verdier, Des catégories dérivées des catégories abéliennes, Astérisque No.
239 (1996), xii+253 pp. (1997).
1. Introduction
It is well-known that many basic constructions from homotopy theory extend to
categories of C∗ -algebras. As we argued in [17], the framework of triangulated
categories is ideal for this purpose. The notion of triangulated category was
introduced by Jean-Louis Verdier to formalise the properties of the derived
category of an Abelian category. Stable homotopy theory provides further clas-
sical examples of triangulated categories. The triangulated category structure
encodes basic information about manipulations with long exact sequences and
(total) derived functors. The main point of [17] is that the domain of the Baum–
Connes assembly map is the total left derived functor of the functor that maps
a G-C∗ -algebra A to K∗ (G r A).
236
Homological algebra in bivariant K-theory 237
(1.4)
238 Ralf Meyer and Ryszard Nest
Definition 4. Let I ⊆ T be an ideal. Its quotient category C/I has the same
objects as C and morphism groups C(A, B)/I(A, B).
equivariance.
If F is a set of closed subgroups of G, we define an ideal VC F in KKG by
The authors feel more at home with Kasparov theory than with spectra. Many
readers will prefer to work in categories of spectra of, say, G-CW-complexes.
242 Ralf Meyer and Ryszard Nest
We do not introduce these categories here; but it shoud be clear enough that
they support similar restriction functors, which provide analogues of the ideals
VC F .
G : KKG → KK, A
→ G A,
We only study these ideals for compact G. In this case, the Green–Julg Theo-
rem identifies K∗ (G A) with the G-equivariant K-theory KG ∗ (A) (see [11]).
Hence the ideal I,K ⊆ KKG is a good equivariant analogue of the ideal IK in
KK.
Literally the same definition as above provides ideals I ⊆ I,K ⊆ KKG
if G is a compact quantum group. We will always allow this more gen-
eral situation below, but readers unfamiliar with quantum groups may ignore
this.
for some chain map f ; the maps B → cone(f ) → A are the canonical ones. It
is well-known that this defines a triangulated category for p = 0; the arguments
for p ≥ 1 are essentially the same.
Another classical example is the stable homotopy category, say, of compactly
generated pointed topological spaces (it is not particularly relevant which cate-
gory of spaces or spectra we use). The suspension is (A) := S1 ∧ A; a triangle
is exact if it is isomorphic to a mapping cone triangle
f
A → B → cone(f ) → A
for some map f ; the maps B → cone(f ) → A are the canonical ones.
We are mainly interested in the categories KK and KKG introduced in §2.2.
Their triangulated category structure is discussed in detail in [17]. We are facing
a notational problem because the functor X
→ C0 (X) from pointed compact
spaces to C∗ -algebras is contravariant, so that mapping cone triangles now
have the form
f
A ← B ← cone(f ) ← C0 (R, A)
T(A, ) : T → Ab, B
→ T(A, B)
T( , B) : Top → Ab, A
→ T(A, B)
Observe that
Tn (A, B) = T( −n A, B) ∼
= T(A, n B) ∼
= T−n (A, B).
tum subgroup H of a locally compact quantum group G and the crossed product
functors G , G r : KKG → KK are exact because they preserve mapping
cone triangles.
Y : C → Fun(Cop , Ab), B
→ T( , B).
Hom(Y(B), F ) ∼
= F (B) for all F ∈∈ Fun(Cop , Ab), B ∈∈ T
Remark 19. Freyd’s Theorem shows that Y induces a bijection between (sta-
ble) exact functors Coh(T) → C and (stable) homological functors T → C
because F̄ ◦ Y is homological if F̄ : Coh(T) → C is exact. Hence the notion
of homological functor is independent of the triangulated category structure
on T because the Yoneda embedding Y : T → Coh(T) does not involve any
additional structure. Hence the notion of homological ideal only uses the sus-
pension automorphism, not the class of exact triangles.
All the ideals considered in §2.2 except for ker κ in Example 5 are ker-
nels of stable homological functors or exact functors. Those of the first kind are
homological by definition. If F : T → T is an exact functor between two trian-
gulated categories, then Y ◦ F : T → Coh(T ) is a stable homological functor
with ker Y ◦ F = ker F by Freyd’s Theorem 16. Hence kernels of exact func-
tors are homological as well.
Is any homological ideal the kernel of an exact functor? This is not the
case:
Proposition 20. Let Der(Ab) be the derived category of the category Ab of
Abelian groups. Define the ideal IH in Der(Ab) as in Example 9. This ideal is
not the kernel of an exact functor.
We postpone the proof to the end of §3.1 because it uses the machinery
of §3.1.
It takes some effort to characterise homological ideals because T/I is almost
never Abelian. The results in [4, §2–3] show that an ideal is homological if
and only if it is saturated in the notation of [4]. We do not discuss this notion
here because most ideals that we consider are obviously homological. The
only example where we could profit from an abstract characterisation is the
ideal ker κ in Example 5.
There is no obvious homological functor whose kernel is ker κ because κ is
not a functor on KK. Nevertheless, ker κ is the kernel of an exact functor; the
relevant functor is the functor KK → UCT, where UCT is the variant of KK
that satisfies the Universal Coefficient Theorem in complete generality. This
functor can be constructed as a localisation of KK (see [17]). The Universal
Coefficient Theorem implies that its kernel is exactly ker κ.
– A morphism f in T is
∗ a ker F -phantom map if and only if F (f ) = 0;
∗ ker F -monic if and only if F (f ) is monic;
∗ ker F -epic if and only if F (f ) is epic;
∗ a ker F -equivalence if and only if F (f ) is invertible.
– An object A ∈∈ T is ker F -contractible if and only if F (A) = 0.
– An exact triangle A → B → C → A is ker F -exact if and only if
Similar things happen for the other ideals in §2.2 that are naturally defined
as kernels of stable homological functors.
Remark 23. It is crucial for the above theory that we consider functors that
are both stable and homological. Everything fails if we drop
either assumption
and consider functors such as K0 (A) or Hom Z/4, K∗ (A) .
Proof. Recall that the generalised mapping cone of f is the object C that fits
f
in an exact triangle A → B → C → A. The long exact sequence for this
triangle yields that F (f ) is invertible if and only if F (C) = 0, where F is
some stable homological functor F with ker F = I. Now the second assertion
follows from Lemma 22. Since the generalised mapping cone of 0 → A is A,
the first assertion is a special case of the second one.
– A morphism f ∈ T(A, B) is
∗ a ker F -phantom map if and only if F (f ) = 0;
∗ ker F -monic if and only if F (f ) is (split) monic.
∗ ker F -epic if and only if F (f ) is (split) epic;
∗ a ker F -equivalence if and only if F (f ) is invertible.
– An object A ∈∈ T is ker F -contractible if and only if F (A) = 0.
– An exact triangle A → B → C → A is ker F -exact if and only if the exact
triangle F (A) → F (B) → F (C) → F (A) in T splits.
Homological algebra in bivariant K-theory 253
You may write down a similar list for the ideal I ⊆ KKG of
Example 7.
Lemma 25 allows us to prove that the ideal IH in Der(Ab) cannot be the
kernel of an exact functor:
in Der(Ab). This triangle is IH -exact because the map Z/2 → Z/4 is injective
as a group homomorphism and hence IH -monic in Der(Ab).
Assume there were an exact functor F : Der(Ab) → T with ker F = IH .
Then F (τ ) = 0, so that F maps our triangle to a split triangle and F (Z/4) ∼ =
F (Z/2) ⊕ F (Z/2) by Lemma 25. It follows that F (2 · idZ/4 ) = 2 · idF (Z/4) = 0
because 2 · idF (Z/2) = F (2 · idZ/2 ) = 0. Hence 2 · idZ/4 ∈ ker F = IH , which
is false. This contradiction shows that there is no exact functor F with ker F =
IH .
dn+1 dn dn−1
C• := (· · · → Cn+1 −−→ Cn −
→ Cn−1 −−→ Cn−2 → · · · )
gn fn hn
Kn+1 −
→ Cn −
→ Kn −
→ Kn+1
Such complexes are called I-exact in [1, 4]. This definition is inspired by the
following well-known fact: a chain complex over an Abelian category is exact
if and only if it splits into short exact sequences of the form Kn Cn Kn−1
as in Definition 26.
We prefer another definition of exactness because we have not found a
general explicit criterion for a chain complex to be I-decomposable.
dn fn gn
Cn −
→ Cn−1 −
→ Xn −
→ Cn . (3.1)
gn fn+1
We call C• I-exact in degree n if the map Xn − → Cn −−−→ Xn+1 belongs
to I(Xn , Xn+1 ). This does not depend on auxiliary choices because the exact
triangles in (3.1) are unique up to (non-canonical) isomorphism.
We call C• I-exact if it is I-exact in degree n for all n ∈ Z.
F (dn+1 ) F (dn )
F (Cn+1 ) −−−−→ F (Cn ) −−−→ F (Cn−1 )
is exact at F (Cn ).
−1 F (gn ) F (fn+1 )
−1 F (Xn ) −−−−−→ F (Cn ) −−−−→ F (Xn+1 )
256 Ralf Meyer and Ryszard Nest
show that the range of −1 F (gn ) and the kernel of F (fn+1 ) are equal to the
kernel of F (dn ) and the range of F (dn+1 ), respectively. Hence C• is I-exact in
degree n if and only if ker F (dn ) ⊆ range F (dn+1 ). Since dn ◦ dn+1 = 0, this is
equivalent to ker F (dn ) = range F (dn+1 ).
Example 30. For the ideal IK in KK, Lemma 28 yields that a chain complex C•
over KK is K-exact (in degree n) if and only if the chain complex
of Z/2-graded Abelian groups is exact (in degree n). Similar remarks apply
to the other ideals in §2.2 that are defined as kernels of stable homological
functors.
As a trivial example, we consider the largest possible ideal I = T. This
ideal is defined by the zero functor. Lemma 28 or the definition yield that
all chain complexes are T-exact. In contrast, it seems hard to characterise the
I-decomposable chain complexes, already for I = T.
∼ α ∼ β ∼ γ (3.2)
A
f
/B g
/C
Homological algebra in bivariant K-theory 257
Remark 32. Lemma 31 implies that I-exact chain complexes of length 3 are
I-decomposable. We do not expect this for chain complexes of length 4. But
we have not searched for a counterexample.
Which chain complexes over T are I-exact for I = 0 and hence for any
homological ideal? The next definition provides the answer.
· · · → −1 A → −1 B → −1 C → A → B → C → A → B → C → · · ·
· · · → −1 C → A → B → C → A → B → C → 2 A → · · ·
Proof. The main issue here is the truncation of chain complexes. Let C• be a
chain complex over T. We embed the map d0 in an exact triangle C0 → C−1 →
X → C0 and let C•≥0 be the chain complex
in Der.
We also have to check that there is no non-zero morphism C• → D• in
Der if C• ∈∈ Der≥0 and D• ∈∈ Der≤−1 . Recall that morphisms in Der are
∼
represented by diagrams C• ← C̃• → D• in Ho(T), where the first map is
an I-equivalence. Hence C̃• ∈∈ Der≥0 as well. We claim that any chain map
f : C̃•≥0 → D•≤−1 is homotopic to 0. Since the maps C̃•≥0 → C• and D• →
D•≤−1 are I-equivalences, any morphism C• → D• vanishes in Der.
It remains to prove the claim. In a first step, we use that D•≤−1 is contractible
in degrees ≥ 0 to replace f by a homotopic chain map supported in degrees <
0. In a second step, we use that C̃•≥0 is homologically exact in the relevant
degrees to recursively construct a chain homotopy between f and 0.
Any truncation structure gives rise to an Abelian category, its core. The core
of Der is the full subcategory C of all chain complexes that are I-exact except
in degree 0. This is a stable Abelian category, and the standard embedding
T → Ho(T) yields a stable homological functor F : T → C with ker F = I.
260 Ralf Meyer and Ryszard Nest
The notions of projective and injective object are dual to each other: if
we pass to the opposite category Top with the canonical triangulated category
structure and use the same ideal Iop , then this exchanges the roles of projective
and injective objects. Therefore, it suffices to discuss one of these two notions
in the following. We will only treat projective objects because all the ideals
in §2.2 have enough projective objects, but most of them do not have enough
injective objects.
Notice that the functor F is I-exact if and only if the associated stable
functor F∗ : T → CZ is I-exact because I is stable.
Homological algebra in bivariant K-theory 261
An I-exact functor also has the following properties (which are strictly
weaker than being I-exact):
Proof. The first assertion follows because direct sums and products of Abelian
groups are exact; the second one is a special case.
Notation 42. Let P be a set of objects of T. We let (P)⊕ be the smallest class
of objects of T that contains P and is closed under retracts and direct sums (as
far as they exist in T).
Proposition 44. If I has enough projective objects, then any object of T has
an I-projective resolution (and vice versa).
Let P• → A and P• → A be I-projective resolutions. Then any map A →
A may be lifted to a chain map P• → P• , and this lifting is unique up to chain
homotopy. Two I-projective resolutions of the same object are chain homotopy
equivalent. As a result, the construction of projective resolutions provides a
functor
P : T → Ho(T).
f g h
Let A → B → C → A be an I-exact triangle. Then there exists a canon-
ical map η : P (C) → P (A)[1] in Ho(T) such that the triangle
P (f ) P (g) η
P (A) −−→ P (B) −−→ P (C) −
→ P (A)[1]
in Ho(T) is exact; here [1] denotes the translation functor in Ho(T), which has
nothing to do with the suspension in T.
P0 @
δ0
/A Pn C
δn
/ Pn−1
@@ CC
@@ CC
f0 @@ f and fn CC fn−1 for n ≥ 1
@ C!
P0 / A Pn / Pn−1
δ0
δn
commute. We must check that this is possible. Since the chain complex P• → A
is I-exact and Pn is I-projective for all n ≥ 0, the chain complexes
(δm )∗ (δ0 )∗
· · · → T(Pn , Pm ) −−→ T(Pn , Pm−1
) → · · · → T(Pn , P0 ) −−→ T(Pn , A) → 0
are exact for all n ∈ N. This allows us to find maps fn as above. By con-
struction, these maps form a chain map lifting f : A → A . Its uniqueness up
to chain homotopy is proved similarly. If we apply this unique lifting result
to two I-projective resolutions of the same object, we get the uniqueness of
I-projective resolutions up to chain homotopy equivalence. Hence we get a
well-defined functor P : T → Ho(T).
Now consider an I-exact triangle A → B → C → A as in the third
paragraph of the lemma. Let X• be the mapping cone of some chain map
P (A) → P (B) in the homotopy class P (f ). This chain complex is supported
in degrees ≥ 0 and has I-projective entries because Xn = P (A)n−1 ⊕ P (B)n .
The map X0 = 0 ⊕ P (B)0 → B → C yields a chain map X• → C, that is, the
composite map X1 → X0 → C vanishes. By construction, this chain map lifts
the given map B → C and we have an exact triangle P (A) → P (B) → X• →
P (A)[1] in Ho(T). It remains to observe that X• → C is I-exact. Then X• is
an I-projective resolution of C. Since such resolutions are unique up to chain
homotopy equivalence, we get a canonical isomorphism X• ∼ = P (C) in Ho(T)
and hence the assertion in the third paragraph.
Let F be a stable homological functor with I = ker F . We have to check that
F (X• ) → F (C) is a resolution. This
reduces
to a well-known
diagram
chase in
Abelian categories, using that F P (A) → F (A) and F P (B) → F (B) are
resolutions and that F (A) F (B) F (C) is exact.
Definition 48. We let ExtnT,I (A, B) be the nth right derived functor with respect
to I of the contravariant functor A
→ T(A, B).
We have natural maps T(A, B) → Ext0T,I (A, B), which usually are not
invertible. Lemma 46 yields long exact sequences
are exact for all n ∈ N. This extension of chain complexes yields the desired
long exact sequence.
We list a few more elementary properties of derived functors. We only spell
things out for the left derived functors Ln F : T → C of a covariant functor
F : T → C. Similar assertions hold for right derived functors of contravariant
functors.
The derived functors Ln F satisfy I ⊆ ker Ln F and hence descend to
functors Ln F : T/I → C because the zero map P (A) → P (B) is a chain
map lifting of f if f ∈ I(A, B). As a consequence, Ln F (A) ∼ = 0 if A
is I-contractible. The long exact homology sequences of Lemma 46
show that Ln F (f ) : Ln F (A) → Ln F (B) is invertible if f ∈ T(A, B) is an
I-equivalence.
266 Ralf Meyer and Ryszard Nest
Warning 49. The derived functors Ln F are not homological and therefore
do not deserve to be called I-exact even though they vanish on I-phantom
maps. Lemma 46 shows that these functors are only half-exact on I-exact
triangles. Thus Ln F (f ) need not be monic (or epic) if f is I-monic (or I-epic).
The problem is that the I-projective resolution functor P : T → Ho(T) is not
exact – it even fails to be stable.
Remark 50. The derived functors introduced above, especially the Ext func-
tors, can be interpreted in terms of derived categories.
We have already observed in §3.2.1 that the I-exact chain complexes
form a thick subcategory of Ho(T). The augmentation map P (A) → A of
an I-projective resolution of A ∈∈ T is a quasi-isomorphism with respect to
this thick subcategory. The chain complex P (A) is projective (see [12]), that
is, for any chain complex C• , the space of morphisms A → C• in the derived
category Der(T, I) agrees with [P (A), C• ]. Especially, ExtnT,I (A, B) is the
space of morphisms A → B[n] in Der(T, I).
Now let F : T → C be an additive covariant functor. Extend it to an exact
functor F̄ : Ho(T) → Ho(C). It has a total left derived functor
LF̄ : Der(T, I) → Der(C), A
→ F̄ P (A) .
By definition, we have Ln F (A) := Hn LF̄ (A) .
Remark 51. In classical Abelian categories, the Ext groups form a graded ring,
and the derived functors form graded modules over this graded ring. The same
happens in our context. The most conceptual construction of these products
uses the description of derived functors sketched in Remark 50.
Recall that we may view elements of ExtnT,I (A, B) as morphisms A → B[n]
in the derived category Der(T, I). Taking translations, we can also view them
as morphisms A[m] → B[n + m] for any m ∈ Z. The usual composition in the
category Der(T, I) therefore yields an associative product
Thus
n we get a graded additive category with morphism spaces
ExtT,I (A, B) n∈N .
Similarly, if F : T → C is an additive functor and LF̄ : Der(T, I) → Der(C)
is as in Remark 51, then a morphism A → B[n] in Der(T, I) induces a mor-
phism LF̄ (A) → LF̄ (B)[n] in Der(C). Passing to homology, we get canonical
Homological algebra in bivariant K-theory 267
maps
ExtnT,I (A, B) → HomC LFm (A), LFm−n (B) ∀m ≥ n,
for all B ∈ T. Let C be the full subcategory of all objects A ∈∈ C for which
F ( (A) is defined. Then F ( is a functor C → T, which we call the (partially
defined) left adjoint of F . Although one usually assumes C = C , we shall also
need F ( in cases where it is not defined
everywhere.
The functor B
→ C A, F (B) for A ∈∈ C vanishes on I = ker F for triv-
ial reasons. Hence F ( (A) ∈∈ T is I-projective. This simple observation is
surprisingly powerful: as we shall see, it often yields all I-projective objects.
where Z[1] means Z in odd degree. Hence K(∗ (Z) = C and K(∗ (Z[1]) = C0 (R).
More generally, Remark 52 shows that K(∗ (A) is defined if both the even and
odd parts of A ∈∈ AbZ/2 are countable free Abelian groups: it is a direct sum
of at most countably many copies of C and C0 (R). Hence all such countable
direct sums are IK -projective (we briefly say K-projective). As we shall see,
K(∗ is not defined on all of AbZ/2 ; this is typical of homological functors.
The reason for the notation PCi is that for a homological functor Fi we usually
take PCi to be the class of projective objects of Ci ; if Fi is exact, then we often
take PCi = Ci . But it may be useful to choose a smaller category, as long as it
satisfies the second condition above.
Proposition 55. In this situation, there are enough I-projective objects, and PI
is generated by i∈I {Fi( (B) | B ∈ PCi }. More precisely, an object of T is
I-projective if and only if it is a retract of i∈I Fi( (Bi ) for a family of objects
Bi ∈ PCi .
Proof. Let P̃0 := i∈I {Fi( (B) | B ∈ PCi } and P0 := (P̃0 )⊕ . To begin with,
we observe that any object of the form Fi( (B) with B ∈∈ PCi is
ker Fi -projective and hence I-projective because I ⊆ ker Fi . Hence P0 consists
of I-projective objects.
Let A ∈∈ T. For each i ∈ I , there is an epimorphism pi : Bi → Fi (A) with
Bi ∈ PCi . The direct sum B := i∈I Fi( (Bi ) exists. We have B ∈∈ P0 by
construction. We are going to construct an I-epimorphism p : B → A. This
shows that there are enough I-projective objects.
The maps pi : Bi → Fi (A) provide maps p̂i : Fi( (B i ) → A via the
adjointness isomorphisms T(Fi( (Bi ), A) ∼ = Ci Bi , Fi (A) . We let p :=
(
p̂i : Fi (Bi ) → A. We must check that p is an I-epimorphism. Equiva-
lently, p is ker Fi -epic for all i ∈ I ; this is, in turn equivalent to Fi (p) being an
epimorphism in Ci for all i ∈ I , because of Lemma 22 or 25. This is what we
are going to prove.
The identity map on Fi((Bi ) yields a map (
αi : Bi → F(i Fi (B i ) via the
adjointness isomorphism T Fi (Bi ), Fi (Bi ) ∼
( (
= Ci Bi , Fi Fi (Bi ) . Compos-
ing with the map
. /
Fi Fi( (Bi ) → Fi Fi( (Bi ) = Fi (B)
that F ◦ F ( (A) ∼
= A holds for A ∈∈ P C because both conditions are evidently
hereditary for direct sums and retracts.
Theorem 59. In the situation of Theorem 57, the domain of the functor F (
contains PC, and its essential range is PI . The functors F and F ( restrict to
equivalences of categories PI ∼= PC inverse to each other.
An object A ∈∈ T is I-projective if and only if F (A) is projective and
C F (A), F (B) ∼ = T(A, B)
for all B ∈∈ T; following Ross Street [24], we call such objects F -projective.
We have F (A) ∈∈ PC if and only if there is an I-equivalence P → A with
P ∈∈ PI .
The functors F and F ( induce bijections between isomorphism classes of
projective resolutions of F (A) in C and isomorphism classes of I-projective
resolutions of A ∈∈ T in T.
If G : T → C is any (stable) additive functor, then there is a unique
right-exact (stable) functor Ḡ : C → C such that Ḡ ◦ F (P ) = G(P ) for all
P ∈∈ PI .
The left derived functors of G with respect to I and of Ḡ are related by
natural isomorphisms Ln Ḡ ◦ F (A) = Ln G(A) for all A ∈∈ T, n ∈ N. There
is a similar statement for cohomological functors, which specialises to natural
isomorphisms
ExtnT,I (A, B) ∼
= ExtnC F (A), F (B) .
Remark 60. The assumption that idempotents split is only needed to check
that the universal I-exact functor has the properties listed in Theorem 57.
The converse directions of Theorem 57 and Theorem 59 do not need this
assumption.
If T has countable direct sums or countable direct products, then idempotents
in T automatically split by [19, §1.3]. This covers categories such as KKG
because they have countable direct sums.
That is, an object of CZ/p is projective if and only if its homogeneous pieces
are.
Theorem 61. The category Ho(C; Z/p) has enough H-projective objects, and
the functor H∗ : Ho(C; Z/p) → CZ/p is the universal H-exact stable homo-
logical functor. Its restriction to PH provides an equivalence of categories
PH ∼ = PCZ/p . More concretely, a chain complex in Ho(C; Z/p) is H-projective
if and only if it is homotopy equivalent to one with vanishing boundary map
and projective entries.
The functor H∗ maps isomorphism classes of H-projective resolutions of a
chain complex A in Ho(C; Z/p) bijectively to isomorphism classes of projective
resolutions of H∗ (A) in CZ/p . We have
ExtnHo(C;Z/p),IH (A, B) ∼
= ExtnC H∗ (A), H∗ (B) .
Let F : C → C be some covariant additive functor and define
Remark 62. Since the universal I-exact functor is essentially unique, the uni-
versality of H∗ : Der(C; Z/p) → CZ/p means that we can recover this functor
and hence the stable Abelian category CZ/p from the ideal IH ⊆ Der(C; Z/p).
That is, the ideal IH and the functor H∗ : Der(C; Z/p) → CZ/p contain exactly
the same amount of information.
For instance, if we forget the precise category C by composing H∗ with some
faithful functor C → C , then the resulting homology functor Ho(C; Z/p) → C
still has kernel IH . We can recover CZ/p by passing to the universal I-exact
functor.
We compare this with the situation for truncation structures ([3]). These
cannot exist for periodic categories such as Der(C; Z/p) for p ≥ 1. Given the
standard truncation structure on Der(C), we can recover the Abelian category C
as its core; we also get back the homology functors Hn : Der(C) → C for all
n ∈ Z. Conversely, the functor H∗ : Der(C) → CZ together with the grading
on CZ tells us what it means for a chain complex to be exact in degrees
Homological algebra in bivariant K-theory 275
greater than 0 or less than 0 and thus determines the truncation structure.
Hence the standard truncation structure on Der(C) contains the same amount
of information as the functor H∗ : Der(C) → CZ together with the grading
on CZ .
Theorem 63. There are enough K-projective objects in KK, and the universal
K-exact functor is K∗ : KK → AbZ/2
c . It restricts to an equivalence of categories
Z/2
between PK and the full subcategory Abfc ⊆ AbZ/2 c of Z/2-graded countable
∗
free Abelian groups. A separable C -algebra belongs to PK if and only if it
is KK-equivalent to i∈I0 C ⊕ i∈I1 C0 (R) where the sets I0 , I1 are at most
countable.
If A ∈∈ KK, then K∗ maps isomorphism classes of K-projective resolutions
of A in T bijectively to isomorphism classes of free resolutions of K∗ (A). We
have
⎧
⎪
⎪ K∗ (A), K∗ (B) for n = 0;
⎨HomAbZ/2
ExtnKK,IK (A, B) =∼ Ext1 Z/2 K∗ (A), K∗ (B) for n = 1;
⎪
⎪ Ab
⎩0 for n ≥ 2.
group and that subgroups of free Abelian groups are again free. Moreover, free
Z/2
Abelian groups are projective. Hence Abfc is the subcategory of projective
objects in AbZ/2
c and any object G ∈∈ AbZ/2
c has a projective resolution of
Z/2
the form 0 → F1 → F0 G with F0 , F1 ∈∈ Abfc . This implies that derived
Z/2
functors on Abc only occur in dimensions 1 and 0.
Z/2
As in Example 53, we see that K(∗ is defined on Abfc and satisfies
K(∗ Z[I0 ], Z[I1 ] ∼
= C⊕ C0 (R)
i∈I0 i∈I1
if I0 , I1 are countable. We also have K∗ ◦ K(∗ Z[I0 ], Z[I1 ] ∼
= Z[I0 ], Z[I1 ] ,
so that the hypotheses of Theorem 57 are satisfied. Hence there are enough
K-projective objects and K∗ is universal. The remaining assertions follow
from Theorem 59 and our detailed knowledge of the homological algebra in
AbZ/2
c .
F : KK → AbZ/2
c , A
→ K∗ (A ⊗ B)
for some B ∈∈ KK, where ⊗ denotes, say, the spatial C∗ -tensor product. We
claim that the associated right-exact functor AbZ/2
c → AbZ/2
c is
F̄ : AbZ/2
c → AbZ/2
c , G
→ G ⊗ K∗ (B).
Here we use the same graded version of Tor as in the Künneth Theorem ([5]).
F : KK → AbZ/2 , B
→ KK∗ (A, B)
G ∈∈ AbZ/2
c as in Example 64. Therefore,
⎧
⎪
⎨KK∗(A, C) ⊗ K∗ (B) for n = 0;
⎪
∼
Ln F (B) = Tor1 KK∗ (A, C), K∗ (B) for n = 1;
⎪
⎪
⎩0 for n ≥ 2.
Example 67. For the ideal IK ⊆ KK, any object has a K-projective resolution
of length 1 by Theorem 63. The other hypothesis of Theorem 66 holds if and
only if A satisfies the Universal Coefficient Theorem (UCT). The UCT for
KK(A, B) predicts KK(A, B) = 0 if K∗ (B) = 0. Conversely, if this is the case,
then Theorem 66 applies, and our description of ExtKK,IK in Theorem 63 yields
the UCT for KK(A, B) for all B. This yields our claim.
Thus the UCT for KK(A, B) is a special of Theorem 66. In the situa-
tions of Examples 64 and 65, we get the familiar Künneth Theorems for
K∗ (A ⊗ B) and KK∗ (A, B). These arguments are very similar to the original
proofs (see [5]). Our machinery allows us to treat other situations in a similar
fashion.
Proof of Theorem 66. We only write down the proof for homological func-
tors. The cohomological case is dual and contains T( , B) as a special
case.
δ1 δ0
Let 0 → P1 −→ P0 −→ A be an I-projective resolution of length 1 and view
it as an I-exact chain complex of length 3. Lemma 31 yields a commuting
diagram
P1
δ1
/ P0 δ̃0
/ Ã
α
P1
δ1
/ P0 δ0
/ A,
such that the top row is part of an I-exact exact triangle P1 → P0 → Ã → P1
and α is an I-equivalence. We claim that α is an isomorphism in T.
α β
We embed α in an exact triangle −1 B → Ã −
→A− → B. Lemma 24 shows
that B is I-contractible because α is an I-equivalence. Hence T(A, B) = 0
by our assumption on A. This forces β = 0, so that our exact triangle splits:
à ∼
= A ⊕ −1 B. Now we apply the functor T(·, B) to the exact triangle P0 →
P1 → Ã. The resulting long exact sequence has the form
Since coker F∗ (δ1 ) = L0 F∗ (A) and ker F∗ (δ1 ) = L1 F∗ (A), we get the desired
exact sequence. The map L0 F∗ (A) → F∗ (A) is the canonical map induced
by δ0 . The other map F∗ (A) → L1 F∗−1 (A) is natural for all morphisms between
objects with an I-projective resolution of length 1 by Proposition 44.
Ext0T,I (A, B) ∼
= T/I(A, B), Ext1T,I (A, B) ∼
= I(A, B).
More generally, we can construct a natural map I(A, B) → Ext1T,I (A, B) for
any homological ideal, using the I-universal homological functor F : T → C.
We embed f ∈ I(A, B) in an exact triangle B → C → A → B. We get
an extension
* +
F (B) F (C) F (A) ∈∈ Ext1C F (A), F (B)
because this triangle is I-exact. This class κ(f ) in Ext1C F (A), F (B) does
not depend on auxiliary choices because the exact triangle B → C →
A → B is unique ∼
up to isomorphism. Theorem 59 yields ExtT,I (A, B) =
1
1
ExtC F (A), F (B) because F is universal. Hence we get a natural map
G : KKG → KK, A
→ G A,
KKG (τ (A), B) ∼
= KK(A, G B) (5.1)
Lemma 69. There are enough projective objects for I and I,K . We have
Z/2
Since any object of AbZ/2
c is a quotient of one in Abfc , Proposition 55 applies
and yields the assertions about I,K and P,K ; even more, P,K is the closure
Z/2
of τ ◦ K(∗ (Abfc ) under retracts.
= C∗ (G) ⊗ A ∼
G τ (A) ∼
A.
=
take the product in reversed order. For classical compact groups, this ring is
commutative, so that the order of multiplication does not matter; in general,
the reversed-order product is the more standard choice. The following fact is
well-known:
Example 71. If C(G) = C∗red (H ) for a discrete group H , then Ĝ = H and the
product on Rep(G) = Z[H ] is the usual convolution. Thus Rep(G) is the group
ring of H .
If G is a compact group, then Rep(G) is the representation ring of G in the
usual sense.
For any B ∈∈ KKG , the Kasparov product turns KKG ∗ (τ C, B) into a left
op ∼
module over the ring KKG0 (τ C, τ C) = Rep(G). Thus KK ∗ (τ C, B) becomes
G
Z/2
an object of the Abelian category Mod(Rep G)c of Z/2-graded countable
Rep(G)-modules. We get a stable homological functor
∗ (τ C, B) is
By (5.1), the underlying Abelian group of KKG
∼ ∼
∗ (τ C, B) = KK∗ (C, G B) = K∗ (G B).
KKG
ε∈{0,1},i∈Iε
Hence the adjoint functor FK( is defined on countable free modules. Idempotents
in KKG split by Remark 60. Therefore, the domain of FK( is closed under
Z/2
retracts and contains all projective objects of Mod(Rep G)c . It is easy to
see that FK ◦ FK( (A) ∼= A for free modules. This extends to retracts and hence
holds for all projective modules (compare Remark 58). Now Theorem 57 yields
that FK is universal.
The assertions about projective objects, projective resolutions, and Ext now
follow from Theorem 59. Theorem 59 also yields a formula for left derived
Z/2
functors in terms of the right-exact functor H̄ : Mod(Rep G)c → C associ-
ated to a homological functor H : KKG → C. It remains to compute H̄ .
First we define the Tor objects in the statement of the theorem if C is the
category of Abelian groups. Then H∗ (τ C) ∈∈ Mod(Rep G)Z/2 , and we can
take the derived functors of the usual Z/2-graded balanced tensor product
⊗Rep G for Rep(G)-modules. We claim that there are natural isomorphisms
H̄∗ (M) ∼
= H∗ (τ C) ⊗Rep G M
Z/2
for all M ∈∈ Mod(Rep G)c . This holds for M = Rep G and hence for all
free modules because we have natural isomorphisms
H̄∗ (Rep G) ∼
= H∗ FK( (Rep G) ∼= H∗ (τ C) ∼
= H∗ (τ C) ⊗Rep G Rep G.
Its derived functors are TornRep G . As above, we see that this yields the derived
functors of H .
The case of cohomological functors is similar and left to the reader.
= KKG (τ C, τ C) → KKG (τ C ⊗ A, τ C ⊗ A) ∼
$A : Rep(G)op ∼ = KKG (τ A, τ A)
Homological algebra in bivariant K-theory 285
for A ∈∈ KK, whose range commutes with the range of the map
τ : KK(A, A) → KKG (τ A, τ A).
The ring KKG (τ A, τ A) acts on KKG (τ A, B) ∼
= KK(A, G B) on the right
by Kasparov product. Hence so does Rep(G)op via $A . Thus KK(A, G B)
becomes a left Rep(G)-module for all A ∈∈ KK, B ∈∈ KKG . These module
structures are natural in the variable A because the images of KKG (τ C, τ C)
and KK(A, A) in KKG (τ A, τ A) commute. Hence they must come from a ring
homomorphism
μB : Rep(G) → KK(G B, G B).
These ring homomorphisms are natural because the Rep(G)-module structures
on KK(A, G B) are manifestly natural in B. Thus we have lifted G to a
functor
F : KKG → KK[Rep G], B
→ (G B, μB ).
It is clear that ker F = I . The target category KK[Rep G] is neither tri-
angulated nor Abelian. To remedy this, we use the Yoneda embedding
Y : KK → Coh(KK) constructed in §2.4. This embedding is fully faithful;
so is the resulting functor KK[Rep G] → Coh(KK)[Rep G].
Theorem 73. The functor Y ◦ F : KKG → Coh(KK)[Rep G] is the universal
I -exact functor.
Proof. We omit the proof of this theorem because it is only notationally more
difficult than the proof of Theorem 72.
The category Coh(KK)[Rep G] is not as terrible as it seems. We can usually
stay within the more tractable subcategory KK[Rep G], and many standard
techniques of homological algebra like bar resolutions work in this setting.
This often allows us to compute derived functors on Coh(KK)[Rep G] in more
classical terms.
Recall that the bar resolution of a Rep G-module M is a natural free resolu-
tion
· · · → (Rep G)⊗n ⊗ M → (Rep G)⊗n−1 ⊗ M → · · · → Rep G ⊗ M → M
with certain natural boundary maps. Defining
(Rep G)⊗n ⊗ M = Z[Ĝn ] ⊗ M := M,
x∈Ĝn
we can make sense of this in C[Rep G] provided C has countable direct sums;
the Rep G-module structures and the boundary maps can also be defined.
286 Ralf Meyer and Ryszard Nest
here HHn (R; M) denotes the nth Hochschild cohomology of a ring R with
coefficients in an R-bimodule M, and KK(G A, G B) is a bimodule over
Rep G via the Kasparov product on the left and right and the ring homomor-
phisms
d
0 → R ⊗2 → R ⊗2 → R, where d(f ) = t · f · t −1 − f . We get
HH0 (R; M) ∼
= HH1 (R; M) ∼
= coker(λρ −1 − 1),
∼ HH0 (R; M) ∼
HH1 (R; M) = = ker(λρ −1 − 1),
and HHn (R; M) = ∼ HHn (R; M) = ∼ 0 for n ≥ 2. Transporting this kind of reso-
lution to C[R], we get that any object of C[R] has an I -projective resolution
of length 1. This would fail for I,K because the category of R-modules has a
non-trivial Ext2 .
The crucial point is that I -equivalences are invertible in KKT . By
Baaj–Skandalis duality, this is equivalent to the following statement: if
f ∈∈ KKZ (A, B) becomes invertible in KK, then it is already invertible
in KKZ . We do not want to discuss here how to prove this. Taking this for
granted, we can now apply Theorem 66 to all objects of KKT .
We write down the resulting exact sequences for KKZ (A, B) for
A, B ∈∈ KKZ because this equivalent setting is more familiar. The actions
of Z on A and B provide two actions of Z on KK∗ (A, B). We let
tA , tB : KK∗ (A, B) → KK∗ (A, B) be the actions of the generators. Theorem 66
yields an exact sequence
coker tA tB−1 − 1|KK∗+1 (A,B) KKZ∗ (A, B) ker tA tB−1 − 1|KK∗ (A,B) .
This is equivalent to a long exact sequence
tA tB−1 −1 tA tB−1 −1
KK1 (A, B) o KKZ1 (A, B) o KK0 (A, B).
References
[1] Javad Asadollahi and Shokrollah Salarian, Gorenstein objects in triangulated
categories, J. Algebra 281 (2004), no. 1, 264–286. MR 2091971
[2] Saad Baaj and Georges Skandalis, C ∗ -algèbres de Hopf et théorie de Kasparov
équivariante, K-Theory 2 (1989), no. 6, 683–721 (French, with English summary).
MR 1010978
[3] Alexander A. Beı̆linson, Joseph Bernstein, and Pierre Deligne, Faisceaux pervers,
Analysis and topology on singular spaces, I (Luminy, 1981), Astérisque, vol. 100,
Soc. Math. France, Paris, 1982, pp. 5–171 (French). MR 751966
288 Ralf Meyer and Ryszard Nest
[24] Ross Street, Homotopy classification of filtered complexes, J. Austral. Math. Soc.
15 (1973), 298–318. MR 0340380
[25] Jean-Louis Verdier, Des catégories dérivées des catégories abéliennes, Astérisque
(1996), no. 239, xii+253 pp. (1997) (French, with French summary). With a preface
by Luc Illusie; Edited and with a note by Georges Maltsiniotis. MR 1453167
[26] Roland Vergnioux, KK-théorie équivariante et opérateurs de Julg-Valette pour les
groupes quantiques, Ph.D. Thesis, Université Paris 7 – Denis Diderot, 2002. elec-
tronically available at http://www.math.jussieu.fr/theses/2002/vergnioux/these.
dvi.
Contents
0. Introduction 291
1. Historical overview 294
2. Background on RHom complexes 301
3. Dualizing complexes 313
4. When Rf∗ respects compacts 325
5. Where we can prove Conjecture 4.16 337
6. Dualizing complexes and f ! 342
7. Several recent results 344
Appendix A. A fact concerning strongly dualizable objects 344
References 346
290
Derived categories and Grothendieck duality 291
0. Introduction
The bulk of this survey will give a modern account of dualizing complexes. Let
X be a noetherian, separated scheme. Let Db (Coh/X) be the bounded derived
category of coherent sheaves on X. We define a dualizing complex to be an
object I ∈ Db (Coh/X) for which the functor RHom(−, I) yields an equivalence
The traditional reason why people cared about dualizing complexes was the
way they behave in the relative situation, where two schemes are involved; we
will discuss this a little more fully in §6. For now we confine ourselves to stating
our relevant new results. Suppose therefore that we are given a morphism f :
X −→ Y , where X and Y are noetherian, separated schemes. Let D(Qcoh/X)
and D(Qcoh/Y ) be the unbounded derived categories of quasicoherent sheaves
on X and Y respectively. There is a pushforward functor Rf∗ : D(Qcoh/X) −→
D(Qcoh/Y ), and it has a left adjoint Lf ∗ and a right adjoint f ! . The next results
are
1 The usual definition of dualizing complexes imposes the hypothesis that I have finite injective
dimension. There is a theorem of Gabber which says that the objects we call “dualizing
complexes” agree with what, in the classical literature, would go under the name “strongly
pointwise dualizing complexes”; see [6, Lemma 3.1.5]. In the classical language, one of the
things we do in this article is show that these complexes have all the good global properties of
the traditional dualizing complexes.
292 Amnon Neeman
Facts 0.3. Hypotheses (a) and (b) of Fact 0.2(ii) hold in either of the following
two situations:
Remark 0.4. I do not, at the present time, fully understand the technical
condition (∗) in Fact 0.3(ii). Conjecture 4.16 articulates the hope that (∗) holds
for all X. We rephrase this slightly: assuming that the technical Conjecture 4.16
holds for every X, then f ! takes dualizing complexes to dualizing complexes as
long as Rf∗ : D(Qcoh/X) −→ D(Qcoh/Y ) takes perfect complexes to perfect
complexes.
Given that at present we do not have Conjecture 4.16 in this generality, in
addition to the preservation of perfect complexes we currently need to assume
that Conjecture 4.16 holds for X. In the remainder of the introduction I will
discuss the status of Conjecture 4.16; there are many Xs for which it is known.
Until now all our schemes were assumed noetherian. Dualizing complexes
are traditionally about producing equivalences Db (Coh/X) ∼
op
= Db (Coh/X),
Derived categories and Grothendieck duality 293
and the category Db (Coh/X) does not obviously make sense unless X is noethe-
rian, or at least coherent. But it turns out that Conjecture 4.16, as well as the
technical condition (∗) in Fact 0.3(ii), make sense much more generally. For the
next result we only assume that our schemes are quasicompact and separated.
In the article we will prove the following assertion
In Remark 0.6 we sketched the argument leading from the more general
Fact 0.5 to Corollary 0.7, valid in the case of noetherian schemes. The reader
should note that Fact 0.3(ii) is immediate from Corollary 0.7. Corollary 0.7
proves more than we need; in Fact 0.3(ii) we only asserted that f ! takes
Db (Coh/Y ) to D+
Coh (Qcoh/X).
Facts 0.8. It remains to review what we know about Conjecture 4.16. The
current state of knowledge is that X satisfies the conjecture if
This finishes the main new results we will present; it remains to give a
brief sketch of the structure of the article. We begin with a fairly extensive
historical review of Grothendieck duality. Then we cover the basic properties
of RHom(−, −); nothing here is especially new, but our approach is to define
RHom(E, −) as the right adjoint of the functor − L ⊗ E, by appealing to Brown
representability. It is therefore a little interesting to see that the theory can
be developed quickly and painlessly; this approach is relatively new, it first
appeared in Murfet’s thesis [48, Appendix C]. Then the rest of the article is
devoted to proving the claims we have made in the Introduction. The one
comment that might be helpful is that our approach hinges on systematically
using products in the categories D(Qcoh/X). These categories are compactly
generated, and therefore have products; see [52, Proposition 8.4.6]. Any right
adjoint will preserve these products, and the functors RHom(E, −), Rf∗ and
f ! are all right adjoints. The idea is to exploit this.
If X is arbitrary then products in the category D(Qcoh/X) are disgusting;
we only really understand them when X is affine. As we will see, this is often
enough. One can frequently reduce oneself to the affine case.
1. Historical overview
Triangulated categories came into being in the early 1960s; they arose inde-
pendently, and more or less simultaneously, in the work of Puppe [55] and of
Verdier [67]2 . Puppe’s interest in the subject came from homotopy theory; we
will say more, much later in this historical account, about the role homotopy
theory played in the early development of the subject. Our survey of the early
history will follow in the footsteps of Verdier; we will explain the problem
that inspired him, and the progress that followed. Verdier came to the subject
2 The reader should note that Verdier’s thesis was only published posthumously, many years after
it was written; the publication date of [67] is misleading.
Derived categories and Grothendieck duality 295
Remark 1.2. Perhaps we should explain the notation. The line bundle #n is
the bundle of holomorphic n–forms on X, and Hom(V, #n ) is the vector bundle
of whose sections, on an open set U ⊂ X, are the holomorphic maps of vector
bundles V|U −→ #n |U . There is an obvious pairing
ϕ
H i (X, V) ⊗ H n−i X , Hom(V, #n ) −−−−→ H n (#n ),
The question that interested people at the time was whether there is a relative
version. Suppose we are given a holomorphic map of complex manifolds f :
X −→ Y . Is there a reasonable general theorem which, in the special case where
Y = {∗} is the one-point space, comes down to Theorem 1.1? The answer turns
out to be Yes, at least in the algebro-geometric framework. In keeping with
tradition I will now switch from the complex analytic setting to the world of
algebraic geometry; I do this because complex analysis has technical difficulties
which I do not want to address.
Let X be a smooth, n–dimensional projective variety over a field k, and let
V be an algebraic vector bundle on X. Serre duality, in its algebraic variety
296 Amnon Neeman
Extn−i (V , W ⊗k #n ).
Now the vector space W can be thought of as a vector bundle over the one-point
space Y = {∗}, and, if you squint hard enough, the isomorphism above begins
to look like an adjunction. We have a functor Rf∗ , which takes a vector bundle
V on X to a string of vector bundles on Y , namely the H i (X, V). And we have
the functor f ! taking the vector bundle W on Y to W ⊗k #n . If we are willing
to treat the Ext’s as Hom’s and to disregard the confusing indices i and (n − i),
it looks like we have an isomorphism
Remark 1.4. Right from the start people disliked derived categories. Derived
categories were quite unlike the more familiar objects of homological algebra,
their behaviour was poorly understood, and people felt uncomfortable using
them. The consensus was that there had to be a better, more natural substitute
for them. People felt that derived categories couldn’t possibly be the real answer,
they couldn’t be the right framework for all these theorems. To this day the
attitude persists; there are many people who are still trying, today, to find a more
natural foundation for this branch of homological algebra. A huge amount of
exciting work has come out in the last few years. The potential replacements
for derived categories include DG–categories, A∞ –categories, stable model
categories, Segal spaces, quasicategories and triangulated derivators.
Because people never liked the formalism, derived categories were slow to
catch on. Now, almost fifty years later, we can point to the impressive theorems
that have been proved using them, and to the extent to which they have come to
permeate far-flung branches of mathematics. As I have already said, the other
chapters in this book contain a compelling case for their success. At this point
we can safely say that, for a theory that everyone badmouthed from the very
outset, derived categories have come a long way.
In Remark 1.4 we mentioned that people have always had a distaste for
derived categories. It might help if we explain one of the features that no one
likes. We begin with
Remark 2.1. Instead of D(Qcoh/X) we could look at DQcoh (X), the derived
category whose objects are complexes of sheaves of OX –modules with qua-
sicoherent cohomology. When X is separated and quasicompact, the obvious
3 Once again the publication dates are misleading; the later paper [50] was published before [4],
which took longer to be accepted. With [50] I happened to be lucky; Thomason was the referee,
liked the result and accepted it immediately. Also, [50] and [51] were written at about the same
time, but published four year apart; [51] took forever to be accepted.
302 Amnon Neeman
We know that the category D(Qcoh/X) is compactly generated; see [51, Propo-
sition 2.5]. From [51, Theorem 4.1] it follows that the coproduct-preserving
triangulated functor − L ⊗ F must have a right adjoint; there is a functor
RHom(F, −) : D(Qcoh/X) −→ D(Qcoh/X), with
Hom E L ⊗ F , G ∼= Hom E , RHom(F, G) .
Reminder 2.2. Next we remind the reader of a technical observation which may
be found in [48, Lemma 6.8]. For this observation we need a little more notation.
4 It is slightly subtle to show that the internal Hom is triangulated in the first variable. For details
see Murfet’s thesis [48, Theorem C.1]. There are also older proofs in the literature, for example
[44, §1.5.3]. Unlike the older arguments, the proof in [48] is in the spirit of this section; it
appeals to the definition of the internal Hom as right adjoint to the tensor product, and develops
the property of the tensor product that would suffice to formally deduce that the right adjoint
must be triangulated in the first variable.
Derived categories and Grothendieck duality 303
The scheme X is still assumed quasicompact and separated, but now we wish
to also consider a quasicompact open subset U ⊂ X. We will let j : U −→ X
stand for the inclusion. Because we now have two schemes, we distinguish
RHomU (F, G) from RHomX (F, G); if F, G are objects of D(Qcoh/U ) then the
first is the RHom which makes sense, while if F, G belong to D(Qcoh/X), then
the second is well-defined.
The remark we want is that, if F ∈ D(Qcoh/X) and G ∈ D(Qcoh/U ), then
there is a natural isomorphism
RHomX (F, Rj∗ G) ∼
= Rj∗ RHomU (j ∗ F, G).
The proof is easy; the result follows from the isomorphisms
HomX E , RHomX (F, Rj∗ G) ∼ = HomX E L ⊗ F , Rj∗ G
∼
= HomU j ∗ E L ⊗ j ∗ F , G
∼
= HomU j ∗ E , RHomU (j ∗ F, G)
∼
= Hom E , Rj∗ RHom (j ∗ F, G) .
X U
HomX E , Rj∗ RHomU (j ∗ F, j ∗ G)
If either (i) or (ii) holds, then the morphism γ (F, G) : j ∗ RHomX (F, G) −→
RHomU (j ∗ F, j ∗ G) of Remark 2.3 is an isomorphism.
Proof. Assume first that both U and X are affine. In that case we have rings
R and S so that X = Spec(R) and U = Spec(S), and we know that the open
immersion j : U −→ X is flat, meaning that S must be a flat R–algebra. The
derived category D(Qcoh/X) identifies with D(R–Mod), the derived category
D(Qcoh/U ) is D(S–Mod), and the functor j ∗ is nothing more than tensoring a
chain complex of R–modules with S. The lemma is now an easy consequence
of the observation that, if M and N are R–modules with M finitely presented,
then the S–module homomorphism
is an isomorphism.
Next we will see how to deduce the case where only U is assumed affine.
We note that, if we fix F and let G vary, the maps γ (F, G) assemble to a natural
transformation between triangulated functors. To show that the maps γ (F, G)
are isomorphisms, for fixed F and all G, it therefore suffices to prove the special
case where G belongs to some class of generators for D(Qcoh/X). The class
we will use is the objects {jVX }∗ G = R{jVX }∗ G, where jVX : V −→ X is the
open immersion of an open affine subset V ⊂ X. Observe that, since V ⊂ X is
assumed affine, the functor {jVX }∗ is exact; hence {jVX }∗ G = R{jVX }∗ G. See [48,
Corollary 3.14] for the fact that these generate.
Because we now have several open subsets involved, we will adopt the
convention that jWW12 : W1 −→ W2 stands for the open immersion between two
subsets W1 ⊂ W2 of X.
The lemma now follows from the isomorphisms
∗ ∗ ∗
{jUX } RHom F , {jVX }∗ G ∼= {jUX } {jVX }∗ RHom {jVX } F , G
∗ ∗
∼
= {jUU∩V }∗ {jUV ∩V } RHom {jVX } F , G
∗ ∗ ∗
∼
= {jUU∩V }∗ RHom {jUV ∩V } {jVX } F , {jUV ∩V } G
∗ ∗ ∗
∼
= {jUU∩V }∗ RHom {jUU∩V } {jUX } F , {jUV ∩V } G
∗ ∗
∼
= RHom {jUX } F , {jUU∩V }∗ {jUV ∩V } G
∗ ∗
∼
= RHom {j X } F , {j X } {j X } G .
U U V ∗
Derived categories and Grothendieck duality 305
The first isomorphism is by Reminder 2.2, the second is by base change, the
third is by the affine case which we already know, the fourth is trivial, the fifth
is again by Reminder 2.2, and finally the sixth is yet another base change.
Finally we need to see how to deduce the general case, where neither U nor
X need be affine. We wish to show that the map γ (F, G) : j ∗ RHomX (F, G) −→
RHomU (j ∗ F, j ∗ G) is an isomorphism. The problem is local; it suffices to show
that it restricts to an isomorphism on each affine open set V ⊂ U . Choose such
an affine open set, and let i : V −→ U be the inclusion. It suffices to show that
the map
i ∗ γ (F, G) : i ∗ j ∗ RHomX (F, G) −−−−→ i ∗ RHomU (j ∗ F, j ∗ G)
is an isomorphism. But we already know that the inclusions i : V −→ U and
j i : V −→ X satisfy the Lemma, since V is affine. Hence both complexes are
naturally isomorphic to RHomV (i ∗ j ∗ F, i ∗ j ∗ G).
Remark 2.5. Suppose X is a quasicompact, separated scheme. Given
two complexes E, F ∈ D(Qcoh/X) we have a counit of adjunction
ε : E L ⊗ RHom(E, F) −→ F. If we have two more complexes E , F ∈
D(Qcoh/X) they also give a map ε : E L ⊗ RHom(E , F ) −→ F . Combin-
ing the two we obtain a single morphism
ε L⊗ ε
E L ⊗ RHom(E, F) L ⊗ E L ⊗ RHom(E , F ) −−−−→ F L ⊗ F ,
and adjunction produces for us a map
RHom(E, F) L ⊗ RHom(E , F ) −−−−→ RHom(E L ⊗ E , F L ⊗ F ).
The special case where E = OX will interest us particularly; in this case we
have a morphism
μ(E, F, F ) : RHom(E, F) L ⊗ F −−−−→ RHom(E , F L ⊗ F ).
We prove:
Lemma 2.6. Let X be a quasicompact, separated scheme. If E ∈ D(Qcoh/X)
is a perfect complex, and F and F in D(Qcoh/X) are arbitrary, then the
morphism μ(E, F, F ) of Remark 2.5 is an isomorphism.
Proof. We have a globally defined morphism in D(Qcoh/X) and wish to prove
it an isomorphism; it suffices to show that, for any open affine subset U ⊂ X,
the morphism μ(E, F, F ) restricts to an isomorphism on U . Let j : U −→ X
be the inclusion. Lemma 2.4 permits us to identify
j ∗ RHom(E, F) ∼ = RHom(j ∗ E, j ∗ F) ,
j ∗ RHom(E, F L ⊗ F ) ∼
= RHom(j ∗ E , j ∗ F L ⊗ j ∗ F )
306 Amnon Neeman
5 The objects we call “strongly dualizable” are labeled “finite” in [40]. In the second last
paragraph on [40, p. 120] the authors refer to [10] for the older terminolgy, which is the one in
current use. Similar notions appeared even earlier, under different names; see the “⊗–categories
rigides” of [57, §I.5, p. 78], or the “compact closed categories” of [33, pp. 102–103].
Derived categories and Grothendieck duality 307
The fact that (c) implies (a) seems new; in Theorem A.1 we will give a
self-contained proof that (a), (b) and (c) are equivalent.
In subsequent sections, especially §4, we will feel free to use (i), (ii) and (iii)
above.
We will find useful the next little lemma; note that, from now until the end of
the section, all our schemes are assumed noetherian.
is an isomorphism in D(Qcoh/U ).
At this point we use hypothesis (i) and Lemma 2.4. Because F ∈
D (Coh/X), we have that j ∗ RHom(F, G) = RHom(j ∗ F, j ∗ G), because E ∈
b
ν(j ∗ E, j ∗ F, j ∗ G) : j ∗ E L ⊗ RHom(j ∗ F, j ∗ G) −→
RHom RHom(j ∗ E, j ∗ F), j ∗ G
is an isomorphism in D(R–Mod).
With our choices of complexes, that is where F and G have been replaced by
K–injective resolutions by injectives and where E is a single, finitely generated
module in degree 0, we obtain a chain map of chain complexes
α : E ⊗ Hom(F, G) −−−−→ Hom Hom(E, F), G .
Do not confuse α with the map ν(E, F, G); we are not yet asserting that they
agree. We will now prove that the morphism α is an isomorphism of chain
complexes. To see this, choose a finite presentation of the module E. That is,
choose an exact sequence
Hom Hom(R m , F) , G / Hom Hom(R n , F) , G / Hom Hom(E, F) , G /0
the Lemma in the special case M = R; we have a finite module R, and (i)
tells us that H i (H) = H i (R ⊗ H) vanishes for i 0. Choose a K–projective
resolution of G −→ H. The map G −→ H is a quasi-isomorphism, and the
complex G may be chosen to belong to K− (R–Proj); that is we can take G to be
a bounded above complex of projective R–modules. Complete the morphism
G −→ H to a distinguished triangle in K(R–Flat)
Our induction will be to show that if (ii) is true for n then (iii) is true for n + 1,
and if (iii) is true for n then (ii) is true for n. Note that (iii) is obviously true for
n = 0; this starts the induction.
M ⊗ Hq ∼
= M L ⊗ Hq ∼
= M L ⊗ 0 = 0.
This gives the acyclicity of M ⊗ Hq .
Notation 2.12. It remains to prove that if (iii) is true for n then so is (ii). We fix
a prime ideal p of height n; we want to prove that Hp is K–flat. Let us simplify
the notation a little. Observe first that X ⊗R Hp ∼ = Xp ⊗Rp Hp ; replacing R
by Rp we may assume R is a local ring of height n, with maximal ideal p
and residue field k = R/p. Let us further replace H by the chain complex Hp .
What we know so far is:
Step 3. With the conventions of Notation 2.12, there exists an integer so that
H i (M ⊗ H) vanishes for all finite R–modules M and for all i > .
3. Dualizing complexes
Throughout this section we continue to assume, as in the second half of §2, that
X is a noetherian, separated scheme. In §2 we prepared the ground by setting up
some technical apparatus, and now we are ready to treat dualizing complexes.
Let us first remind the reader of the definition.
E
→ RHom(E, I)
gives an equivalence
Remark 3.2. Perhaps we should remind the reader: when E and I are two
objects of Db (Coh/X), the complex RHom(E, I) has no obligation to belong
to Db (Coh/X); there is no reason to expect its cohomology to be bounded
above. It might help to recall the special case when X = Spec(R) is affine, and
Db (Coh/X) reduces to Db (R–mod). Consider two finite R–modules M and N ;
they are objects of Db (R–mod), and RHom(M, N ) is a chain complex whose
cohomology is Extn (M, N). It is perfectly possible to have Extn (M, N )
= 0
for infinitely many n.
As the above example illustrates it is a restriction on I to demand that, for
every E, the object RHom(E, I) be isomorphic to an object in Db (Coh/X). It
is a severe restriction to further insist that the functor
RHom(−, I) : Db (Coh/X) −−−−→ Db (Coh/X)
op
is its own left adjoint, and the unit and counit of adjunction are the restrictions
to Db (Coh/X) of the unit and counit of adjunction on D(Qcoh/X).
Now general category theory kicks in and tells us that, if G is a functor with
a left adjoint F , then G will be an equivalence if and only if both the unit and
the counit of adjunction are isomorphisms. In our particular case, where the
unit and counit happen to agree, this comes down to the following.
Proof. To deduce Proposition 3.6 from Lemma 3.5, we need to show that it
suffices to check that the morphism
E −−−−→ RHom RHom(E, I), I
Remark 3.8. The next obvious question is what happens if we are given,
on the same scheme X, two dualizing complexes I and J. How do they
compare?
It is clear that, if we are given a dualizing complex I, an integer n and a line
bundle L, then n L L ⊗ I is also a dualizing complex; tensoring with a line
bundle and suspending is harmless. It is also harmless to suspend by different
integers on different connected components of X. The remarkable fact is that
this is all the freedom we have. Up to these basic moves, the dualizing complex
is unique. We state this as a lemma and give the proof.
Lemma 3.9. Let I and J be two dualizing complexes on the same scheme X.
Then I = L L ⊗ J, where L is locally some suspension of a line bundle. On each
connected component Xi ⊂ X there is an integer ni , a line bundle Li , and an
isomorphism L|Xi ∼= ni Li .
J = RHom(O X , J)
= RHom RHom(I, I), J by Proposition 3.6(ii)
= I L ⊗ RHom(I, J) by Lemma 2.10
= I L⊗ L by definition of L .
ηi DI ηj D I
1 −−−−→ DI DI −−−−−→ DI DJ DJ DI
Derived categories and Grothendieck duality 317
Hk (L) are coherent sheaves on X, all but finitely many of which vanish. It
follows that there exists a Zariski open set U ⊂ X so that, on U , the coherent
sheaves Hk (L) satisfy
0 if k
= −n
Hk (L)|U =
L if k = −n
where L is some line bundle on U . Now the integer n is locally constant, and
must be constant on connected components. On each connected component Xi
there is an integer ni so that Hk (L) = 0 for k
= −ni , and H−ni (L) is a line
bundle. This makes L|Xi a complex whose cohomology sheaves vanish except
in only one dimension, and hence it must be isomorphic in Db (Coh/Xi ) to
ni H−ni (L).
Remark 3.10. Let I be a dualizing complex. It is an object of Db (Coh/X), and
has a bounded-below injective resolution. In the homotopy category K(Inj/X),
whose objects are chain complexes of injective quasicoherent sheaves on X,
there is a bounded-below complex I and a quasi-isomorphism I −→ I . Fur-
thermore, I is unique up to homotopy. If J is another dualising complex,
Lemma 3.9 informs us that J = L L ⊗ I, with L locally isomorphic to a shift of
a line bundle. Obviously, L ⊗ I is an injective resolution for J.
While the injective resolutions I ∈ K(Inj/X) of dualizing complexes I are
only determined up to twisting by complexes L, the Hom–complexes Hom(I, I )
are objects of K(Flat/X) well defined up to homotopy; we have
Hom(I, I ) ∼
= Hom(L ⊗ I, L ⊗ I ).
Even though the dualizing complex I started its life as an object of the derived
category Db (Coh/X), subject to conditions that appear innocuous enough, we
have learned
(i) The object I is unique up to tensor by an L, with L locally isomorphic to
a shift of a line bundle.
(ii) If we replace I by its injective resolution, then the object Hom(I, I) is a
well-defined complex of flat OX –modules, unique up to homotopy.
In Proposition 3.6 we learned that the natural map OX −→ Hom(I, I) must be
an isomorphism in Db (Coh/X). It is a homology isomorphism of complexes
of flat modules, but not usually a homotopy equivalence.
Our next observation is
Theorem 3.11. Being a dualizing complex is local. In other words let Y = ∪Ui
be an open cover. Then a complex I ∈ D(Qcoh/Y ) is dualizing if and only if,
for each i, the restriction of I to Ui is dualizing.
Derived categories and Grothendieck duality 319
Let us begin with (i). The proof that (b)=⇒(a) is easy. Let E belong to
Db (Coh/Y ); by Lemma 2.4 we know that
By (b) the term on the left is in Db (Coh/Ui ) for every i, and from the iso-
morphism with the term on the right we conclude that RHom(E, I) must be in
Db (Coh/Y ).
Slightly subtler is (a)=⇒(b). Suppose (a) holds, and Ei is an object of
Db (Coh/Ui ). Since coherent sheaves and morphisms of coherent sheaves can
be extended from the open set Ui to all of Y , the object Ei ∈ Db (Coh/Ui ) is
isomorphic to fi∗ G, where G ∈ Db (Coh/Y ). Therefore
Remark 3.13. In Remark 3.12 we noted that dualizing complexes are preserved
when we pull them back by open immersions. It is natural to wonder what
happens with other functors.
320 Amnon Neeman
Remark 3.14. Before we state the theorem, it might help if we remind the
reader of what boundedness properties are always, unconditionally true, for
any morphism f : X −→ Y ; this might help separate what is formal from the
genuinely restrictive hypotheses of the theorem.
We are assuming that X and Y are noetherian and separated. This means that
the functor Rf∗ can be computed using Čech cohomology with respect to an
affine open cover of X, and this cover may be taken finite. If E ∈ D(Qcoh/X)
is bounded it follows that Rf∗ E will also be a bounded complex. That much is
free. But we have no obvious way to control the size (as in coherence versus
quasicoherence) of the finitely many cohomology sheaves of Rf∗ E.
It turns out to be a strong restriction on f to assume that Rf∗ :
D(Qcoh/X) −→ D(Qcoh/Y ) respects the bounded, coherent subcategories.
In symbols: it may happen that Rf∗ will take the subcategory Db (Coh/X) ⊂
D(Qcoh/X) to the subcategory Db (Coh/Y ) ⊂ D(Qcoh/Y ), but it certainly does
not come for free. The usual sufficient condition is that the map f : X −→ Y
be a proper morphism of finite type. This concludes what we will need to know
concerning the functor Rf∗ , to make sense of the statement of Theorem 3.15;
the preservation of boundedness is formal, the preservation of coherence is not.
Next we want to look at the properties of the functor f ! which play a role
in the statement of Theorem 3.15. It helps to begin by returning briefly to
the functor Rf∗ , and observing a little more closely the precise bounds that
the Čech complex gives. Suppose that X admits a cover by n + 1 affine open
sets; then the Čech complex computing Rf∗ has length n. If E is an object in
D<m (Qcoh/X), that is E ∈ D(Qcoh/X) is quasi-isomorphic to a complex van-
ishing in degrees ≥ m, then Rf∗ E must be in D<m+n (Qcoh/Y ); a Čech complex
of length n can only raise the cohomological degree by at most n. This means
that for any F ∈ D≥m+n (Qcoh/Y ) we have Hom(Rf∗ E, F) = 0. Adjunction
gives us Hom(E, f ! F) = 0, and this is true for all E ∈ D<m (Qcoh/X). Hence
Derived categories and Grothendieck duality 321
f ! F ∈ D≥m (Qcoh/X). We have just shown that the functor f ! must take
bounded-below complexes to bounded-below complexes. In particular the
image under f ! of the subcategory Db (Coh/Y ) ⊂ D(Qcoh/Y ) must lie in
D+ (Qcoh/X) ⊂ D(Qcoh/X).
That ends our free ride. There is no formal reason to expect the func-
tor f ! to take Db (Coh/Y ) ⊂ D(Qcoh/Y ) into Db (Coh/X), or even into the
larger D+ Coh (Qcoh/X). We remind the reader: the objects in the category
D+Coh (Qcoh/X) are the bounded-below chain complexes of quasicoherent
sheaves, whose cohomology is all coherent.
Next we prove:
Remark 3.16. The reader might wish to compare Theorem 3.15 with older
results in the literature; see, for example, [15, Remark 2, p. 299], [66, Corol-
lary 3, p. 396], [2, Proposition 2.5.11] and [45, §9.1 and §9.2].
Reminder 3.17. In the proof of Theorem 3.15 we will appeal to the technical
result [46, Theorem 4.2]; let us therefore remind the reader. The technical result
asserts the following: in the category D(Qcoh/X) there is a compact generator
S which detects non-vanishing high cohomology. Precisely, this means that
there is an integer A = A(S), depending only on S, so that
Perhaps we should explain the notation: the Hk means the kth cohomology
sheaf of the chain complex of sheaves. The result informs us that we can
tell whether the sheaf cohomology vanishes, above a certain degree, just by
computing some groups, namely Hom(S, k G).
(iv) For all objects G ∈ Db (Coh/X) we will show that RHom(G, f ! I) belongs
to Db (Coh/X). The case where G = OX will prove that f ! I ∈ Db (Coh/X).
(v) We will show that the natural map OX −→ RHom(f ! I, f ! I) is an isomor-
phism.
Let us begin with (iv). What we are given is that G belongs to Db (Coh/X),
and hypothesis (ii) of the Theorem says that f ! I must belong to D+
Coh (Qcoh/X).
+
It follows that RHom(G, f I) has to be in DCoh (Qcoh/X). We remind the reader:
!
the complex G is finite, and hence we can immediately reduce to the case where
G is a single coherent sheaf concentrated in degree 0. If G is a single sheaf,
then there is a spectral sequence converging to the cohomology sheaves of the
ij ij
complex RHom(G, f I), whose E2 term is Ext G , H (f I) . The sheaves E2
! i j !
is equivalent. And the point is that we can compute the groups on the right;
they are
Hom S , RHom(G , k f ! I) = Hom(S ⊗ GL, f I) k
L k !
also belong to Db (Coh/Y ). Its cohomology may be computed using the Čech
complex on Y , and hence is bounded. Therefore
Hom Rf∗ (S L ⊗ G), k I = H k RHom Rf∗ (S L ⊗ G), I
finite type. All we observed concerning (ii) was that it is not automatic; we
made no mention of any interesting examples of f s which satisfy (ii). It is time
to remedy this. We begin with the easy
Remark 4.3. The fact that Lf ∗ respects products means that it must have a left
adjoint. Van den Bergh
* ! suggested+ a formula for this adjoint; it should be given
by the functor Rf∗ f OY ⊗ − . Lipman and Van den Bergh independently
L
found proofs that the formula works, at least for large classes of f s.
is an isomorphism.
f !Y ∼
= f ! OY L ⊗ Lf ∗ Y ,
Remark 4.6. While f ! OY need not in general be compact, there are interesting
things one can say about it, facts which I do not fully understand. To illustrate
one of the strange features, a phenomenon which seems mysterious to me,
consider the following. By [51, Theorem 5.4] there is an isomorphism f ! (−) ∼ =
f ! OY L ⊗ Lf ∗ (−). Applying Rf∗ to this isomorphism, and using the projection
formula, we deduce an isomorphism
* +
Rf∗ f ! (−) ∼
= Rf∗ f ! OY L ⊗ − .
The functor on the left is the composite of two right adjoints, hence commutes
with products. It follows that so does the functor on the right; Reminder 2.7(iii)
now informs us that Rf∗ f ! OY has to be compact. Even though f ! OY need not
be compact, its pushforward Rf∗ f ! OY must be.
In the light of Lemma 4.4 and Remark 4.5 it makes sense to study the class
of objects L ∈ D(Qcoh/X) for which the functor L L ⊗ − commutes with the
products of Lemma 4.4. Let us make this a definition.
Remark 4.8. When X can be understood from context we will omit it from
the notation; that is we will write S for S(X). Reminder 2.7(iii) tells us that
all compact objects belong to S = S(X). Lemma 4.4 says that, if f : X −→ Y
is a morphism where Rf∗ takes compacts to compacts, then f ! OY belongs
to S(X). Clearly, if E −→ F −→ G −→ E is a distinguished triangle, and
if two of the objects belong to S, then so does the third. Not so trivial is the
lemma
f L⊗ 1
But E is compact; any map E −→ G L ⊗ H factors as E −→ F L ⊗ H −→
G L ⊗ H, with F ∈ D(Qcoh/X) compact. Applying this to the above we deduce
a factorization
! " ! "
f L⊗ 1
E −−−−→ F ⊗ L
Xλ −−−−→ L ⊗ L
Xλ −−−−→ (L L ⊗ Xλ ).
λ∈ λ∈ λ∈
330 Amnon Neeman
Fi YYYYYY fi
YYYYYY
YYYY,
ϕi
eeeeee2 L
eeeee eeee
F e fi+1
i+1
Derived categories and Grothendieck duality 331
to an isomorphism; the reader can find this many places, for example in [51,
§3]. Since E generates it follows that ϕ is an isomorphism; L is isomorphic to
- fi , a countable homotopy colimit of compact objects. Hence L is
Hocolim
ℵ1 –compact. This establishes (i).
It remains to prove (ii); from now we assume E is a strong generator, and our
notation will be as in [5, §2.2]. Let n be an integer so that En = {D(Qcoh/X)}c .
Recall our sequence above, of morphisms fi : Fi −→ L. One can show, by an
induction on i which we leave to the reader,
already vanishes.
Proof. We apply Lemma 4.9 with E = R. We discover that there must exist
a single compact object F, that is a bounded complex of finitely generated,
projective R–modules, as well single map f : F −→ L, so that any morphism
n R −→ L factors through f . The morphisms n R −→ L are in bijection
with elements of H −n (L) and we learn that, for every i, the map H i (F) −→
H i (L) must be surjective. But F is bounded, hence H i (L) vanishes for all but
finitely many i. It remains to show that L is isomorphic in D(R–Mod) to a
bounded above chain complex of finitely generated projectives.
Suppose that i is the integer at which the cohomology of L stops; that
is H i (L)
= 0, but H j (L) = 0 for all j > i. It suffices to produce a finitely
generated, projective module P i and an epimorphism P i −→ H i (L). Let us
first establish that this really suffices. Suppose such a morphism exists; it
must lift to a map in the derived category −i P i −→ L = L0 . The third
edge of the triangle −i P i −→ L −→ L1 −→ −i+1 P i is an object L1 ∈
S(X) whose cohomology stops at (i − 1). We can now apply induction to
obtain a sequence of finitely generated projectives and morphisms −k P k −→
Lk , where the cohomology of Lk stops at i − k, and these finitely generated
projectives assemble to a chain complex
· · · −→ P i−2 −→ P i−1 −→ P i −→ 0
quasi-isomorphic to L.
It therefore remains to prove the existence of the surjection P i −→ H i (L).
Let P be a K–projective resolution for L; we may choose P to be a bounded
above complex of projective modules, which stops at i. That is P is a chain
complex
· · · −→ Pi−2 −→ Pi−1 −→ Pi −→ 0
and H i (P) ∼
= H i (L)
= 0. The morphism f : F −→ P is a map in the derived
category between bounded above chain complexes of projectives, and hence
may be realized by a chain map. That is, we have a chain map
∂ i+1
F
· · · −−−−−−→ F i−2 −−−−−−→ F i−1 −−−−−−→ F i −−−−−−→ F i+1 −−−−−−→ F i+2 −−−−−−→ · · ·
⏐ ⏐ ⏐ ⏐ ⏐
⏐ ⏐ ⏐ ⏐ ⏐
From the first paragraph of the proof we know that the map H i (F) −→ H i (P)
is surjective; that is the kernel of ∂Fi+1 surjects onto the cokernel of ∂Pi . But then
Fi certainly must also surject to the cokernel; we have found our surjection
Fi −→ H i (P), with Fi finitely generated and projective.
334 Amnon Neeman
This ends what I can currently prove in glorious generality. Let me next
formulate a conjecture.
Remark 4.15. In view of Lemma 4.12, Conjecture 4.13 amounts to the state-
ment that if L ∈ S(X) and j : U −→ X is an open immersion of an affine open
subset, then j ∗ (L) is in S(U ).
(∗) For any quasicompact open subset U ⊂ X there exist a compact object
E ∈ D(Qcoh/X), and an integer n ≥ 1, so that
Rj∗ OU ∈ En ;
Reminder 4.17. Perhaps we should remind the reader of the notation in [5,
Definition 2.2.3]. Let E be an object in some triangulated category T. The
full subcategory E1 is defined to be the one containing all direct summands
of all coproducts of arbitrary suspensions of E. The category En is defined
inductively; an object lies in En+1 if it is a direct summand of an object y,
and y fits in a triangle
Remark 4.18. Now that we have reminded the reader of the notation in Con-
jecture 4.16, we should also explain its relevance. In the remainder of this
section we will show that Conjecture 4.16 implies Conjecture 4.13. In §5 we
will study the many cases in which we know Conjecture 4.16 to be true. And
then, in §6, we will return to the relation with Grothendieck duality.
Derived categories and Grothendieck duality 335
All three facts are easy; we leave (i) and (ii) to the reader and indicate the proof
of (iii). Let E be a compact object, L an object in S, and assume all the tλ lie in
T . We have isomorphisms
4 5 ! "
L ⊗L
(E ⊗ t ) =
L ∼ L ⊗E ⊗
L L
t Reminder 2.7(iii)
λ λ
λ∈ 4 λ∈ 5
∼
= E L⊗ (L L ⊗ tλ ) Because L ∈ S, all tλ ∈ T
λ∈
∼
= (L L ⊗ E L ⊗ tλ ) Reminder 2.7(iii) .
λ∈
Lemma 4.21. Conjecture 4.16 implies Conjecture 4.13. More precisely for
each X, Conjecture 4.16 for X implies Conjecture 4.13 for X. We spell this
out: if X is a quasicompact, separated scheme, and if Conjecture 4.16 holds for
X, then every object L ∈ S(X) is pseudocoherent, and the cohomology sheaves
Hi (L) vanish for all but finitely many i.
Now take an object L ∈ S(X) and a set of objects {tλ , λ ∈ }, where each tλ
belongs to T , and observe the string of isomorphisms
4 ! "5 ! "
∗ L
j∗ j L ⊗ tλ ∼
= L ⊗ j∗
L
tλ projection formula
λ∈ ! λ∈ "
∼
= L L⊗ j∗ t j∗ has left adjoint
λ
λ∈
∼
= (L L ⊗ j∗ tλ ) j∗ T is adapted
λ∈
∼
= j∗ (j ∗ L L ⊗ tλ ) projection formula
λ∈4 5
∼
= j∗ ∗
(j L ⊗ tλ )
L
j∗ has left adjoint .
λ∈
(i) U is quasicompact.
(ii) Let j : U −→ X be the inclusion. There exist a compact object E ∈
D(Qcoh/X), and an integer n ≥ 1, so that Rj∗ OU ∈ En .
6 The reader should note that the statement of [5, Theorem 3.1.4] is slightly different; the result
we want is merely a step in the proof. See the last paragraph of [5, §3.4] for details.
338 Amnon Neeman
Proof. The open sets Xf , f ∈ R form a basis for the topology of X = Spec(R);
any open set U is the union of the Xf ’s contained in it. If U is quasicompact
we may express it as a finite union of Xf ’s. We will prove, by induction on n,
(i) The union of n basic open sets, that is open sets of the form Xf , is a decent
open set.
Lemma 5.4 gives us the case n = 1. Suppose therefore that we know (i) for n,
that U = ∪ni=1 Xfi is the union of n open sets Xfi , and that V = Xg is another
basic open set. We want to show that U ∪ V is decent, and Lemma 5.5 tells us
that it suffices to show that U , V and U ∩ V are. The case of V = Xg comes
from Lemma 5.4, while for U and
! n "
n
n
U ∩V = Xfi ∩ Xg = (Xfi ∩ Xg ) = Xfi g
i=1 i=1 i=1
we appeal to the fact that both can be covered by n basic open sets.
F L ⊗ R{j2 }∗ OU m ⊂ F L ⊗ Gmn .
Proof. We are assuming that X is quasicompact and has a cover by decent open
affines. There must exist a finite subcover. We may cover X by n decent open
affines, for some integer n ≥ 1; the proof will be by induction on n. The case
n = 1 comes from Lemma 5.6. Suppose therefore that we know the assertion
for schemes X which admit covers by n decent open affines, and let X be
a scheme admitting a cover by n + 1 decent open affines. Then X = U ∪ V ,
where U has a cover by n decent open affines and V is a decent open affine.
Lemma 5.8 informs us that U ⊂ X is decent. Let W ⊂ X be any quasicompact
open subset; we want to prove that W is decent.
To do this observe that W can be written as the union
W = (W ∩ U ) ∪ (W ∩ V ) .
Now X is assumed separated, and hence the intersection of any two quasicom-
pact open sets is quasicompact; we deduce that W ∩ U , W ∩ V and W ∩ U ∩ V
are all quasicompact. Induction tells us that
Derived categories and Grothendieck duality 341
Lf ∗ R{jVY }∗ OV ∼
= R{jfX−1 V }∗ Lf ∗ OV ∼
= R{jfX−1 V }∗ Of −1 V .
Remark 5.12. There are two useful situations where Corollary 5.11 applies.
They are:
compute
( )
HomY (Rf∗ E, F) ∼
= HomY Rf∗ E , RHomY RHomY (F, I) , I
* +
∼
= HomY Rf∗ E L ⊗ RHomY (F, I), I
( )
∼
= HomY Rf∗ E L ⊗ Lf ∗ RHomY (F, I) , I
* +
∼
= HomX E L ⊗ Lf ∗ RHomY (F, I), f ! I
( )
∼
= HomX E , RHomX Lf ∗ RHomY (F, I) , f ! I ;
the first isomomorphism is by Lemma 3.5(ii), the second because RHom is right
adjoint to the tensor, the third by the projection formula, the fourth because
f ! is right adjoint to Rf∗ , and the fifth by the adjunction between tensor and
RHom. Put together, we deduce a natural isomorphism
f !F ∼
= RHomX Lf ∗ RHomY (F, I) , f ! I . (††)
Lemma 4.4 informs us that f ! OY belongs to S(X), and, using (i) and (ii) above
as well as Lemma 4.21, we conclude that f ! OY is in D− (Coh/X). We therefore
deduce
(iv) f ! takes Db (Coh/Y ) to Db (Coh/X); see Remark 0.6 and Corollary 0.7.
(v) If I is a dualizing complex on Y , then f ! I is a dualizing complex on X;
see Fact 0.3(ii).
If we look at (iii) and (iv), they tell us that the adjoint pair of functors (Rf∗ , f ! )
restrict to functors between the subcategories Db (Coh/X) ⊂ D(Qcoh/X) and
344 Amnon Neeman
f ! S = f ! OY L ⊗ Lf ∗ S
tells us, at least in principle, how to work out all there is to know about the
functor f ! : D(Qcoh/Y ) −→ D(Qcoh/X).
This approach is very new, poorly understood, and in my opinion it has great
promise. Limitations of space prevent me from discussing it any further here.
Assume that the unit of the tensor is compact, and that all compact objects are
strongly dualizable. If E ∈ T is some object, then the following are equivalent:
(i) E is compact.
(ii) E is strongly dualizable.
(iii) Tensor product with E commutes with arbitrary products in T; that is, the
natural map
E∧ tλ −−−−→ (E ∧ tλ )
λ∈ λ∈
The fact that it respects products means that it must be representable; see [52,
Theorem 8.6.1]. There is an object G ∈ D(Qcoh/X) and a natural isomorphism
References
[1] Leovigildo Alonso Tarrı́o, Ana Jeremı́as López, and Joseph Lipman, Local homol-
ogy and cohomology on schemes, Ann. Sci. École Norm. Sup. (4) 30 (1997), no. 1,
1–39.
[2] , Studies in duality on Noetherian formal schemes and non-Noetherian
ordinary schemes, American Mathematical Society, Providence, RI, 1999.
[3] Leovigildo Alonso Tarrı́o, Ana Jeremı́as López, and Marı́a José Souto Salorio,
Localization in categories of complexes and unbounded resolutions, Canad. J.
Math. 52 (2000), no. 2, 225–247.
[4] Marcel Bökstedt and Amnon Neeman, Homotopy limits in triangulated categories,
Compositio Math. 86 (1993), 209–234.
Derived categories and Grothendieck duality 347
[5] Alexei I. Bondal and Michel van den Bergh, Generators and representability of
functors in commutative and noncommutative geometry, Mosc. Math. J. 3 (2003),
no. 1, 1–36, 258.
[6] Brian Conrad, Grothendieck duality and base change, Lecture Notes in Mathe-
matics, vol. 1750, Springer-Verlag, Berlin, 2000.
[7] Pierre Deligne, Cohomology à support propre en construction du foncteur f ! ,
Residues and Duality, Lecture Notes in Mathematics, vol. 20, Springer–Verlag,
1966, pp. 404–421.
[8] , Cohomologie à supports propres, Théorie des topos et cohomologie étale
des schémas. Tome 3 (Berlin), Lecture Notes in Mathematics, vol. 305, Springer-
Verlag, 1973, (SGA 4, Exposé XVII), pp. 250–461.
[9] Duiliu-Emanuel Diaconescu, Enhanced D-brane categories from string field the-
ory, J. High Energy Phys. (2001), no. 6, Paper 16, 19.
[10] Albrecht Dold and Dieter Puppe, Duality, trace, and transfer, Proceedings of
the International Conference on Geometric Topology (Warsaw, 1978) (Warsaw),
PWN, 1980, pp. 81–102.
[11] Fouad El Zein, Complexe dualisant et applications à la classe fondamentale d’un
cycle, Bull. Soc. Math. France Mém. (1978), no. 58, 93.
[12] Pierre-Paul Grivel, Une démonstration du théorème de dualité de Verdier, Enseign.
Math. (2) 31 (1985), no. 3-4, 227–247.
[13] Alexandre Grothendieck, The cohomology theory of abstract algebraic varieties,
Proc. Internat. Congress Math. (Edinburgh, 1958), Cambridge Univ. Press, New
York, 1960, pp. 103–118.
[14] , Théorèmes de dualité pour les faisceaux algébriques cohérents,
Séminaire Bourbaki, Vol. 4, Soc. Math. France, Paris, 1995, pp. Exp. No. 149,
169–193.
[15] Robin Hartshorne, Residues and duality, Lecture Notes in Mathematics, vol. 20,
Springer–Verlag, 1966.
[16] Glenn W. Hopkins, An algebraic approach to Grothendieck’s residue symbol,
Trans. Amer. Math. Soc. 275 (1983), no. 2, 511–537.
[17] Glenn W. Hopkins and Joseph Lipman, An elementary theory of Grothendieck’s
residue symbol, C. R. Math. Rep. Acad. Sci. Canada 1 (1978/79), no. 3, 169–
172.
[18] Kentaro Hori and Johannes Walcher, D-branes from matrix factorizations, C. R.
Phys. 5 (2004), no. 9-10, 1061–1070, Strings 04. Part I.
[19] Mark Hovey, John H. Palmieri, and Neil P. Strickland, Axiomatic stable homotopy
theory, vol. 128, Memoirs AMS, no. 610, Amer. Math. Soc., 1997.
[20] Reinhold Hübl and Ernst Kunz, Integration of differential forms on schemes, J.
Reine Angew. Math. 410 (1990), 53–83.
[21] , Regular differential forms and duality for projective morphisms, J. Reine
Angew. Math. 410 (1990), 84–108.
[22] Reinhold Hübl and Pramathanath Sastry, Regular differential forms and relative
duality, Amer. J. Math. 115 (1993), no. 4, 749–787.
[23] Luc Illusie, Conditions de finitude, Théorie des intersections et théorème
de Riemann-Roch, Springer-Verlag, Berlin, 1971, Séminaire de Géométrie
Algébrique du Bois-Marie 1966–1967 (SGA 6, Exposé III), pp. 222–273. Lecture
Notes in Mathematics, Vol. 225.
348 Amnon Neeman
[43] , Lectures on local cohomology and duality, Local cohomology and its
applications (Guanajuato, 1999), Lecture Notes in Pure and Appl. Math., vol. 226,
Dekker, New York, 2002, pp. 39–89.
[44] , Notes on derived categories and Grothendieck duality, LNM, Springer-
Verlag, to appear.
[45] Joseph Lipman, Suresh Nayak, and Pramathanath Sastry, Pseudofunctorial behav-
ior of Cousin complexes on formal schemes, Variance and duality for Cousin
complexes on formal schemes, Contemp. Math., vol. 375, Amer. Math. Soc.,
Providence, RI, 2005, pp. 3–133.
[46] Joseph Lipman and Amnon Neeman, Quasi-perfect scheme maps and bound-
edness of the twisted inverse image functor, Illinois J. Math. 51 (2007), 209–
236.
[47] Ralf Meyer and Ryszard Nest, The Baum-Connes conjecture via localisation of
categories, Topology 45 (2006), no. 2, 209–259.
[48] Daniel S. Murfet, The mock homotopy category of projectives and Grothendieck
duality, (PhD thesis, Australian National U. 2008).
[49] Suresh Nayak, Pasting pseudofunctors, Variance and duality for Cousin complexes
on formal schemes, Contemp. Math., vol. 375, Amer. Math. Soc., Providence, RI,
2005, pp. 195–271.
[50] Amnon Neeman, The connection between the K–theory localisation theorem of
Thomason, Trobaugh and Yao, and the smashing subcategories of Bousfield and
Ravenel, Ann. Sci. École Normale Supérieure 25 (1992), 547–566.
[51] , The Grothendieck duality theorem via Bousfield’s techniques and Brown
representability, Jour. Amer. Math. Soc. 9 (1996), 205–236.
[52] , Triangulated Categories, Annals of Mathematics Studies, vol. 148,
Princeton University Press, Princeton, NJ, 2001.
[53] , The homotopy category of flat modules, and Grothendieck duality, Invent.
Math. 174 (2008), 255–308.
[54] , Some adjoints in homotopy categories, Annals of Mathematics Studies,
vol. 171, Princeton University Press, Princeton, NJ, 2010, pp. 2143–2155.
[55] Dieter Puppe, On the structure of stable homotopy theory, Colloquium on algebraic
topology, Aarhus Universitet Matematisk Institut, 1962, pp. 65–71.
[56] Jean-Pierre Ramis, Gabriel Ruget, and Jean-Louis Verdier, Dualité relative en
géométrie analytique complexe, Invent. Math. 13 (1971), 261–283.
[57] Neantro Saavedra Rivano, Catégories Tannakiennes, Springer-Verlag, Berlin,
1972, Lecture Notes in Mathematics, Vol. 265.
[58] Pramathanath Sastry, Base change and Grothendieck duality for Cohen-Macaulay
maps, Compos. Math. 140 (2004), no. 3, 729–777.
[59] , Duality for Cousin complexes, Variance and duality for Cousin complexes
on formal schemes, Contemp. Math., vol. 375, Amer. Math. Soc., Providence, RI,
2005, pp. 137–192.
[60] Jean-Pierre Serre, Un théorème de dualité, Comment. Math. Helv. 29 (1955),
9–26.
[61] Nicolas Spaltenstein, Resolutions of unbounded complexes, Compositio Math. 65
(1988), no. 2, 121–154.
[62] Robert W. Thomason and Thomas F. Trobaugh, Higher algebraic K–theory of
schemes and of derived categories, The Grothendieck Festschrift (a collection of
350 Amnon Neeman
papers to honor Grothendieck’s 60’th birthday), vol. 3, Birkhäuser, 1990, pp. 247–
435.
[63] Michel van den Bergh, Existence theorems for dualizing complexes over non-
commutative graded and filtered rings, J. Algebra 195 (1997), no. 2, 662–679.
[64] Jean-Louis Verdier, Le théorème de dualité de Poincaré, C. R. Acad. Sci. Paris
256 (1963), 2084–2086.
[65] , A duality theorem in the etale cohomology of schemes, Proc. Conf. Local
Fields (Driebergen, 1966), Springer, Berlin, 1967, pp. 184–198.
[66] Jean-Louis Verdier, Base change for twisted inverse images of coherent sheaves,
vol. Collection: Algebraic Geometry, Tata Inst. Fund. Res., 1968, pp. 393–408.
[67] , Des catégories dérivées des catégories abeliennes, Asterisque, vol. 239,
Société Mathématique de France, 1996 (French).
[68] Amnon Yekutieli, Dualizing complexes over noncommutative graded algebras, J.
Algebra 153 (1992), no. 1, 41–84.
[69] , An explicit construction of the Grothendieck residue complex, Astérisque
(1992), no. 208, 127, With an appendix by Pramathanath Sastry.
[70] Amnon Yekutieli and James J. Zhang, Dualizing complexes and perverse modules
over differential algebras, Compos. Math. 141 (2005), no. 3, 620–654.
[71] , Dualizing complexes and perverse sheaves on noncommutative ringed
schemes, Selecta Math. (N.S.) 12 (2006), no. 1, 137–177.
Contents
1. Introduction 351
2. Notations 352
3. Localisation 352
3.1. Abelian case 352
3.2. Triangulated case 354
3.2.1. Derived functors 354
3.2.2. Open subvarieties and quotients 354
3.2.3. Perfect complexes 355
3.2.4. Extensions of perfect complexes 356
3.2.5. Applications to K-theory 359
4. Reconstruction 360
4.1. Abelian case 360
4.1.1. Classification of Serre subcategories 360
4.1.2. Centres 361
4.2. Triangulated case 362
4.2.1. Classification of thick subcategories 362
4.2.2. Centres 363
4.2.3. Remarks on centres 364
4.2.4. Affine varieties 366
4.2.5. (Anti-)ample canonical bundles 366
References 369
1. Introduction
We describe some basic properties of the derived category of coherent sheaves
on a variety (bounded derived category or perfect complexes).
The first chapter considers the problem of extending vector bundles from an
open subset. Thomason and Trobaugh provided an answer to this problem by
351
352 Raphaël Rouquier
2. Notations
We fix a field k and we call variety a separated scheme of finite type over k.
Given a variety X, we denote by X-coh (resp. X-qcoh) the category of coherent
(resp. quasi-coherent) sheaves on X.
All functors between triangulated categories are assumed to be triangulated.
We denote by Z(C) the centre of a category C (=endomorphisms of the
identity functor).
Given a ring R, we denote by R-mod the category of finitely generated
R-modules.
3. Localisation
3.1. Abelian case
Let us recall some basic properties of categories of coherent sheaves.
Let X be a variety. Given Z closed in X, we denote by X-cohZ the full
subcategory of X-coh of coherent sheaves with support contained in Z. This is
a Serre subcategory of X-coh.
Derived categories and algebraic geometry 353
0 → ker F → A → B → 0.
0 → ker F → T → T → 0.
Proof of Theorem 3.11 and Lemma 3.10. Let us show first that the lemma for X
implies the theorem for X. The combination of Lemmas 3.8, 3.9 and 3.10 shows
that j ∗ induces a fully faithful functor X-perf Z /X-perf Z∩Z → U -perf U ∩Z .
Let I be the image of that functor: this is a full triangulated subcategory.
Since X-perf Z = DZ (X-qcoh) (lemma 3.10), we have Ī = DU ∩Z (U -qcoh).
It follows from Lemma 3.9 that U -perf Z ∩U is the thick subcategory generated
by I. The theorem follows now from Lemma 3.7.
Lemma 3.9 shows that Lemma 3.10 will follow from the fact that
(X-perf Y )⊥ = 0.
Let us assume first that X is affine. Let {y1 , . . . , yr } be a family of generators
of the defining ideal of Y and let
6
r
yi
Gr = (0 → OX −
→ OX → 0)
i=1
358 Raphaël Rouquier
be the associated Koszul complex (the non-zero terms are in degrees −r, . . . , 0).
We will show by induction on r that an object C ∈ DY (X-qcoh) vanishes if
Gr ⊗ C = 0. The lemma will follow, since Hom(G∨r , C[i]) H i (Gr ⊗ C),
where G∨r = R Hom(Gr , OX ) is the dual of Gr . The case r = 0 is clear.
Consider r > 0 and C ∈ DY (X-qcoh) a non-zero object. By induction, there
exists i such that H i (Gr−1 ⊗ C)
= 0. The distinguished triangle
yr
Gr−1 ⊗ C −
→ Gr−1 ⊗ C → Gr ⊗ C
yr
gives an exact sequence H i−1 (Gr ⊗ C) → H i (Gr−1 ⊗ C) − → H i (Gr−1 ⊗ C).
Since H (Gr−1 ⊗ C) is supported by the closed subvariety (yr = 0), we deduce
i
Exercice 3.1. Let C ∈ X-qcoh such that for any set I of objects of X-qcoh, the
canonical map D∈I Hom(C, D) → Hom(C, D∈I D) is an isomorphism.
Show that C is coherent.
Derived categories and algebraic geometry 359
Remark 3.13. Let us show, following Serre [Se, 5.a, p.371], that Theorem 3.6
doesn’t hold for vector bundles.
Let X = A3 and U = X − {0}. Let F be the vector bundle on U which is
the pullback of the tangent bundle on P2 . The restriction map K0 (X) → K0 (U )
is an isomorphism. Since F is not the direct sum of two line bundles, it is not
the restriction of a vector bundle on X.
Let G be a coherent sheaf on X extending F. The second syzygy #2 G of G
is locally free (hence free) and this provides a perfect complex extending F :
there is a complex of free sheaves
0 → #2 G → P −1 → P 0 → 0
Remark 3.14. A proof similar to that of Proposition 3.1 shows that there is
an exact sequence 0 → DZb (X-coh) → D b (X-coh) → D b (U -coh) → 0. When
X is smooth, then X-perf = D b (X-coh), hence Theorem 3.6 is a consequence
of that exact sequence. In this case, the canonical map K0 (X) → K0 (U ) is
surjective.
· · · → Ki (X on Z) → Ki (X) → Ki (U ) → Ki−1 (X on Z) → · · ·
Theorem 3.17. Let U and V be open subsets of X. Then there is a long exact
sequence
· · · → Ki (U ∪ V ) → Ki (U ) ⊕ Ki (V ) → Ki (U ∩ V ) → Ki−1 (U ∪ V ) → · · ·
360 Raphaël Rouquier
4. Reconstruction
4.1. Abelian case
We will start with the classical case of coherent sheaves, following Gabriel
[Ga].
4.1.2. Centres
Lemma 4.3. Let R be a ring. The canonical map Z(R) → Z(R-mod) is an
isomorphism.
Corollary 4.4 (Gabriel). The abelian category X-coh determines the variety
X.
Proof. We define a ringed space E. Its points are the irreducible Serre sub-
categories of finite type of X-coh. The open subsets are the D(I), defined
as the set of those subcategories J that are not contained in a given Serre
subcategory I.
Theorem 4.1 shows that the map sending a point x ∈ X to X-coh{x} defines
a homeomorphism X → E.
Consider the presheaf of rings on E given by OE (D(I)) = Z(X-coh /I).
If D(I ) ⊂ D(I), then the quotient functor X-coh /I → X-coh /I induces a
map Z(X-coh /I) → Z(X-coh /I ). We denote by OE the associated sheaf.
The canonical map (U ) → Z(U -coh) induces a morphism of ringed spaces
X → E. To check that this is an isomorphism, it is enough to consider its
restriction to an open affine subset: Lemma 4.3 provides the conclusion in the
affine case.
Remark 4.5. Actually, Lemma 4.3 holds for non-affine varieties: the canonical
map (OX ) → Z(X-coh) is an isomorphism of rings. That follows from the
construction of the category X-coh by gluing the abelian categories Ui -coh
along the quotient categories (Ui ∩ Uj )-coh, given a finite open covering of X
by open affine subsets Ui .
As a consequence, the presheaf in the proof of Corollary 4.4 is a sheaf.
362 Raphaël Rouquier
Lemma 4.7. Let Z be a closed subset of X. Then there exists a perfect complex
on X with support Z.
Proof. Let C ∈ X-perf with support Z and let I be the ideal of X-perf generated
by C. Lemmas 3.9 and 3.10 show that I = X-perf Z if and only if DZb (X-coh) ⊂
Ī.
Derived categories and algebraic geometry 363
Remark 4.9. The classical proof of Theorem 4.6 uses the following result
(“tensor nilpotence Theorem”).
Let C ∈ X-perf, let D ∈ D(X-qcoh) and let f : C → D. We assume that
for every point x of X, we have f ⊗ k(x) = 0 in D(k(x)-Mod).
Then there is an integer n such that ⊗n f : ⊗n C → ⊗n D vanishes in
D(X-qcoh).
4.2.2. Centres
We proceed as in §4.1.2 to obtain a reconstruction Theorem [Ba1, Rou1].
Let X be a variety. We define Z(X-perf)lnil as the subring of Z(X-perf)
given by elements α such that α(C) is nilpotent for all C ∈ X-perf. We put
Z(X-perf)lred = Z(X-perf)/Z(X-perf)lnil .
Remark 4.15. Let k be a field and let C be a k-linear category with finite
dimensional Hom-spaces. Let P be an indecomposable object of C and let
φ ∈ Z(End(P )) with the following properties:
r φ2 = 0
r given x ∈ End(P ) such that φx
= 0 or xφ
= 0, then x is invertible.
Given Q an indecomposable object of C not isomorphic to P , then
r Hom(P , Q)φ Hom(Q, P ) = 0
r for f ∈ Hom(P , Q) non zero, then, φ ∈ Hom(Q, P )f and for g ∈
Hom(Q, P ) non zero, then, φ ∈ g Hom(P , Q).
366 Raphaël Rouquier
φ
Let C = (0 → P − → P → 0), a complex with non-zero terms in degrees 0
and 1. Define an endomorphism ζ of C as φ in degree 1 and 0 elsewhere.
Then, one shows there is a unique element of the centre of the homotopy
category T of complexes (all, bounded, bounded above or bounded below, . . . )
of objects of C with the following properties:
r it is 0 on indecomposable objects of T which are not isomorphic to C[i] for
some i.
r it is φ[i] on C[i].
This applies to C the category of finitely generated projective A-modules
when A = k[x]/(x 2 ) (or A = Z/4Z by slightly modifying the setting above):
the centre of A-perf is larger that A.
Remark 4.16. Proposition 4.14 does not extend to Hochschild cohomology
[Ca]. Let X be an elliptic curve. Then, H H 2 (X) = Ext2X×X (OX , OX )
=
0. On the other hand, X-coh is a hereditary category. In particular,
Hom(Id, Id[2]) = 0, where Id is the identity functor of D b (X-coh).
Proof. The crucial point is the fact that the Serre functor is intrinsic to the
category D b (X-coh). Since the thick subcategories invariant by the Serre func-
tor and its inverse are ideals (cf §4.2.4 and Lemma 4.18), we recover X from
D b (X-coh).
∼
Consider now F : D b (X-coh) → D b (Y -coh). Note that F commutes with
the Serre functors : F SX SY F .
Let Z be a closed subset of Y . Since F −1 (DZb (Y -coh)) is a thick subcategory
of D b (X-coh) stable under SXi for all i, it is of the form D b (Z) (X-coh) and this
provides an injection from closed subsets of Y to closed subsets of X.
Assume Z is irreducible. Let V be an open affine subset of Y . Then
∼
F −1 induces an equivalence D b (V -coh) → D b ((X − (Y − V ))-coh) that
∼
restricts to an equivalence DZb (V -coh) → D b (Z) ((X − (Y − V ))-coh). Since
DZb (V -coh) is an irreducible thick subcategory (§4.2.4 and Theorem 4.6), we
deduce that D b (Z) ((X − (Y − V ))-coh) is an irreducible thick subcategory
of D b ((X − (Y − V ))-coh), hence (Z) ∩ (X − (Y − V )) is irreducible.
If Y = V1 ∪ · · · ∪ Vr is a covering by open affine subsets, then the subsets
X − (Y − Vi ) give an open affine covering of X, and it follows that (Z) is
irreducible.
We define an injection φ : Y → X between points by φ(y) = (y). If y is
b
a closed point, then by Lemma 4.21 below the thick subcategory D{y} (Y -coh)
b
of D (Y -coh) is minimal as a non-zero thick subcategory. It follows that
F −1 (D{y}
b
(Y -coh)) = Dφ(y) (X-coh) is a minimal non-zero thick subcategory
of D (X-coh) that is stable under SXi for all i. We deduce that φ(y) is a closed
b
point.
368 Raphaël Rouquier
X → Y is continuous.
Let U = Y − Z. We have a sequence of canonical isomorphisms (cf
Lemma 4.10)
∼ ∼ ∼
(U ) → Z(D b (U -coh))lred −−→ Z(D b (φ(U )-coh))lred → (φ(U )).
F −1
Remark 4.22. Bondal and Orlov [BoOr] show that the structure of graded
category of D b (X-coh) (we forget the distinguished triangles) is enough to
reconstruct X in Theorem 4.19.
∼
Given α : X → Y an isomorphism of varieties, we have an equivalence
∼
α∗ : D b (X-coh) → D b (Y -coh). Given L a line bundle on X, we have a self-
∼
equivalence L⊗? : D b (X-coh) → D b (X-coh). Finally, given n ∈ Z, we have
the self-equivalence [n].
This gives a injective morphism from Z × (Pic X Aut(X)) to the group
Aut(D b (X-coh)) of isomorphism classes of self-equivalences of D b (X-coh).
We can now complete Theorem 4.19.
Derived categories and algebraic geometry 369
Remark 4.24. A proof similar to the one of Theorem 4.23 shows that Pic(X)
∼
Aut(X) → Aut(X-coh) for any variety X.
References
[Ba1] P. Balmer, Presheaves of triangulated categories and reconstruction of
schemes, Math. Ann. 324 (2002), 557–580.
[Ba2] P. Balmer, The spectrum of prime ideals in tensor triangulated categories,
J. Reine Angew. Math. 588 (2005), 149–168.
[BaFl] P. Balmer and G. Favi, Gluing techniques in triangular geometry, Q. J. Math.
58 (2007), 415–441.
370 Raphaël Rouquier
[BoOr] A. Bondal and D. Orlov, Reconstruction of a variety from the derived category
and groups of autoequivalences, Compositio Math. 125 (2001), 327–344.
[Ca] A. Caldararu, letter to Tom Bridgeland, 5 November 2002.
[Ga] P. Gabriel, Des catgories abéliennes, Bull. Soc. Math. France 90 (1962),
323–448.
[Ho] M. Hopkins, Global methods in homotopy theory, in “Homotopy theory
(Durham, 1985)”, 73–96, London Math. Soc. Lecture Note Ser., 117, Cam-
bridge Univ. Press, 1987.
[KaScha] M. Kashiwara and P. Schapira, “Categories and sheaves”, Springer Verlag,
2006.
[Nee1] A. Neeman, The chromatic tower for D(R), Topology 31 (1992), 519–532.
[Nee2] A. Neeman, The connection between the K-theory localization theorem of
Thomason, Trobaugh and Yao and the smashing subcategories of Bousfield
and Ravenel, Ann. Sci. Ecole Norm. Sup. 25 (1992), 547–566.
[Nee3] A. Neeman, “Triangulated categories”, Princeton University Press, 2001.
[Or] D. Orlov, Equivalences of derived categories and K3 surfaces, J. Math. Sci.
(New York) 84 (1997), 1361–1381.
[Rou1] R. Rouquier, Peccot lectures, Collège de France, March 2000.
[Rou2] R. Rouquier, Dimensions of triangulated categories, Journal of K-theory 1
(2008), 193–256.
[Se] J.-P. Serre, Prolongement des faisceaux analytiques cohérents, Ann. Inst.
Fourier 16 (1966), 363–374.
[Th1] R.B. Thomason, The local to global principle in algebraic K-theory, Pro-
ceedings ICM, Kyoto, 1990, 381–394, Springer Verlag, 1991.
[Th2] R.B. Thomason, The classification of triangulated subcategories, Compositio
Math. 105 (1997), 1–27.
[ThTr] R.B. Thomason and T.F. Trobaugh, Higher algebraic K-theory of schemes
and of derived categories, Grothendieck Festschrift vol. III, 247–435,
Birkhauser, 1990.
[To] B. Toen, The homotopy theory of dg-categories and derived Morita theory,
Invent. Math. 167 (2007), 615–667.
Triangulated categories for the analysts
pi erre s ch api ra
Abstract. This paper aims at showing how the tools of Algebraic Geometry
apply to Analysis. We will review various classical constructions, including Sato’s
hyperfunctions, Fourier-Sato transform and microlocalization, the microlocal
theory of sheaves (with some applications to PDE) and explain the necessity of
Grothendieck topologies to treat algebraically generalized functions with growth
conditions.
Introduction
In this paper, we will show how the tools of Algebraic Geometry– sheaves,
triangulated and derived categories, Grothendieck topologies and stacks– play
(or should play) a crucial role in Analysis.
Note that, conversely, some problems of Analysis led to new algebraic con-
cepts. For example, one of the deepest notion related to triangulated categories
is that of t-structure, and as a particular case, that of perverse sheaves, and
these notions emerged with the study of the Riemann Hilbert correspondence,
a problem dealing with differential equations.
Another example is the Fourier transform, clearly one of the most essential
tools of the analysts, until it was categorified by Sato and applied to algebraic
analysis, next transposed to algebraic geometry (the Fourier-Mukai transform).
The classical analysts are used to work in various functional spaces con-
structed with the machinery of functional analysis and Fourier transform, but
Sato’s construction of hyperfunctions [28] in the 60’s does not use any of these
tools. It is a radically new approach which indeed has entirely modified the
mathematical landscape in this area. The functional spaces are now replaced by
“functorial spaces”, that is, sheaves of generalized holomorphic functions on a
complex manifold X or, more precisely, complexes of sheaves RHom (G, OX ),
where G is an object of the derived category of R-constructible sheaves on the
real underlying manifold to X. Putting general systems of linear partial dif-
ferential equations, i.e., DX -modules, in the machinery, one is led to study
complexes RHom (G, F ), where F = Sol(M) := RHom DX (M, OX ) is the
complex of holomorphic solutions of a coherent DX -module M.
371
372 Pierre Schapira
0.1 Notations
In this paper, we mainly follow the notations of [16].
(i) We denote by k a field and by Db (kX ) the bounded derived category of
sheaves of k-vector spaces on a topological space X. More generally, if A is
a sheaf of rings on X, we denote by Mod(A) the category of left A-modules
and by Db (A) its bounded derived category. If there is no risk of confusion, we
write RHom instead of RHom kX and similarly for RHom and ⊗.
(ii) If Z is a locally closed subset of the topological space X, we denote by
kXZ the sheaf which is the constant sheaf with stalk k on Z and which is 0 on
X \ Z. If there is no risk of confusion, we write kZ instead of kXZ .
(iii) For a real manifold X, we denote by dimX its dimension and by orX the
orientation sheaf. For a morphism of manifolds f : X − → Y , we set orX/Y =
orX ⊗f −1 orY .
(iv) We denote by DX ( • ) = RHom kX ( • , kX ) the duality functor in Db (kX )
and by ωX the dualizing complex. Recall that ωX orX [dimX ].
(v) For a (real or complex) manifold X, we denote by τ : T X − → X and
π : T ∗X − → X its tangent bundle and cotangent bundle, respectively. Let
f:X− → Y be a morphism of (real or complex) manifolds. To f are associated
Triangulated categories for the analysts 373
the maps
TX J
f
/ X ×Y T Y fτ
/ T Y, T ∗ X Lo
fd
X ×Y T ∗ Y
fπ
/ T ∗Y
JJ LLL
JJ LLL π
J
τ JJJ π LLL
τ τ π
J% L%
X
f
/Y X
f
/ Y.
We denote by TX∗ X the zero-section of T ∗ X and by TX∗ Y = fd−1 TX∗ X the conor-
mal bundle to f . If f is an embedding, we identify TX∗ Y to a sub-bundle of
T ∗ Y and call it the conormal bundle to X.
(vi) For a complex manifold X, we denote by OX the sheaf of holomorphic
functions and by #X the sheaf of holomorphic forms of maximal degree. We
denote by dX the complex dimension of X.
1. Generalized functions
In the sixties, people used to work in various spaces of generalized functions
on a real manifold. The situation drastically changed with Sato’s definition of
hyperfunctions [28].
Consider first the case where M is an open subset of the real line R and let
X an open neighborhood of M in the complex line C satisfying X ∩ R = M.
The space B(M) of hyperfunctions on M is given by
It is easily proved that this space depends only on M, not on the choice of X,
and that the correspondence U
→ B(U ) (U open in M) defines a flabby sheaf
BM on M.
Classically, the “boundary value” of a holomorphic function ϕ(z) defined in
the open set X ∩ {Im z > 0} of the complex line, if it exists, is the limit (for a
>
suitable topology) of the function ϕ(x + iy) as y −
→ 0. With Sato’s definition,
the boundary value always exists and is no more a limit. Indeed, it is the
class of the holomorphic function ψ(z) ∈ O(X \ M) given by ψ(z) = ϕ(z) for
Im z > 0 and ψ(z) = 0 for Im z < 0.
On a manifold M of dimension n, the sheaf BM was originally defined as
BM = HM
n
(OX ) ⊗ orM
b : O(#) −
→ B(M). (1.2)
Both examples (1.1) and (1.3) are described by a sheaf of the type
RHom (G, OX ), with G a constructible sheaf (see [16] for an exposition).
Recall that a sheaf G on a real analytic manifold X is R-constructible if there
exists a subanalytic stratification of X on which G is locally constant of finite
rank (over the field k). One denotes by DbR−c (kX ) the full triangulated sub-
category of Db (kX ) consisting of objects with R-constructible cohomology.
On a complex manifold, replacing the subanalytic stratifications by complex
analytic stratifications, one also gets the category DbC−c (kX ) of C-constructible
sheaves.
The advantage of considering the category DbR−c (kX ) is that the properties
of being constructible (in the derived sense) is stable by the six Grothendieck
operations (with suitable properness hypotheses).
To summarize, the classical functional spaces are now replaced by the “func-
torial spaces” RHom CX (G, OX ), where G ∈ DbR−c (CX ).
Triangulated categories for the analysts 375
2. D-modules
References for the theory of D-modules are made to [8, 9].
The theory of D-modules appeared in the 70’s with Kashiwara’s thesis [8]
and Bernstein’s paper [2]. However, already in the 60’s, Sato had the main
ideas of the theory in mind and gave talks at Tokyo University on these topics.
Unfortunately, Sato did not write anything and it seems that his ideas were not
understood at this time. (See [1, 30].)
Let X be a complex manifold. One denotes by DX the sheaf of rings of
holomorphic (finite order) differential operators. A system of linear differential
equations on X is a left coherent DX -module. The link with the intuitive notion
of a system of linear differential equations is as follows. Locally on X, M may
be represented as the cokernel of a matrix ·P0 of differential operators acting
on the right. By classical arguments of analytic geometry (Hilbert’s syzygies
theorem), one shows that, locally, M admits a bounded resolution by free
modules of finite type, that is, M is locally isomorphic to the cohomology of
a bounded complex
• Nr N1 N0 ·P0
M := 0 −
→ DX −
→ ··· −
→ DX −→ DX −
→ 0. (2.1)
N0 N1P0 · Nr
Sol(M) 0 −
→ OX −→ OX −
→ ··· −
→ OX −
→ 0, (2.3)
Operations on D-modules
Let f : X −
→ Y be a morphism of complex manifolds. The sheaf
−1
→Y := OX ⊗f −1 OY f DY
DX− (2.4)
One says that f is non characteristic for N ∈ Dbcoh (DY ) if (see Notations 0.1)
One defines the direct image functor Df ! for left D-modules by using the line
bundles #X and #Y (or their inverse) which intertwine the left and right struc-
tures. If M ∈ Dbgood (DX ) and f is proper on the support of M, one deduces from
Grauert’s direct images Theorem that Df ! M belongs to Dbgood (DY ). Moreover,
there is a natural isomorphism
Holonomic systems
Since the characteristic variety of a coherent DX -module M is involutive,
it is natural to study with a particular attention the extreme case where this
characteristic variety has minimal dimension.
Simple examples in dimension one show that this functor is not fully faithful,
but Kashiwara-Oshima [15] and Kashiwara-Kawai [14] gave the definition of
a “regular holonomic” D-module and Kashiwara [11] proved the Riemann-
Hilbert correspondence, that is, the equivalence of categories
∼
(Dbreg−hol (DX ))op −→ DbC−c (CX )
where now Dbreg−hol (DX ) denotes the full triangulated subcategory of Db (DX )
consisting of objects with regular holonomic cohomology.
Integral transforms
In this subsection, we shall study the action of integral transforms on D-modules
and their sheaves of generalized solutions.
Consider two complex manifolds X and Y , of complex dimension dX and
dY respectively, and the correspondence:
X × YG g
f
ww GGG
{www G#
X Y
L ◦ G = Rf ! (L ⊗ g −1 G). (2.7)
Now we consider
This result was first stated (in a slightly less general formulation) in [4]. Its
proof makes use of the isomorphisms (2.5) and (2.6).
This result admits several variants: one may replace OX with its temperate
version or formal version (see § 5 below and [17, Ch. 7]), there are twisted
versions and also G-equivariant versions (see [13]).
Many applications of this theorem are exposed in [3], in particular to
projective duality, Penrose transform and, more generally, Grasmanniann
correspondences.
3. Microsupport
References for this section are made to [16].
Let X denote a real manifold of class C ∞ and let F ∈ Db (kX ). The micro-
support SS(F ) of F is the closed conic subset of T ∗ X defined as follows.
(Rϕ≥0 (F ))x0 = 0.
Assume now that X is a complex manifold, that we identify with its real
underlying manifold. The link between the microsupport of sheaves and the
characteristic variety of coherent D-modules is given by:
Setting F = Sol(M), we are reduced to study RHom (G, F ). Our only infor-
mation is now purely geometrical, this is the microsupport of G and that of F
(this last one being the characteristic variety of M). Now, we can forget that
we are working on a complex manifold and that we are dealing with partial
differential equations. We are reduced to the microlocal study of sheaves on a
real manifold [16].
Application: ellipticity
Let us show how the classical Petrowsky regularity theorem may be obtained
with the only use of the Cauchy-Kowalevsky-Leray Theorem, and some sheaf
theory.
The regularity theorem for sheaves is as follows.
RHom (G, kX ) ⊗ F −
→ RHom (G, F ) (3.1)
is an isomorphism.
to M if
In other words, the two complexes of real analytic and hyperfunction solutions
of an elliptic system are quasi-isomorphic. This is the Petrowsky’s theorem for
D-modules.
Of course, this result extends to other sheaves of generalized holomorphic
functions, replacing the constant sheaf CM with an R-constructible sheaf G.
For further developments, see [31].
4. Microlocal analysis
For a detailed exposition, see [16].
Fourier-Sato transform
Let τ : E −
→ M be a finite dimensional real vector bundle over a real manifold
M with fiber dimension n, π : E ∗ −→ M the dual vector bundle. Denote by p1
and p2 the first and second projection defined on E ×M E ∗ , and define:
P = {(x, y) ∈ E ×M E ∗ ; x, y ≥ 0}
P = {(x, y) ∈ E ×M E ∗ ; x, y ≤ 0}
E ×M EK∗
p1 tt KKpK2
t K%
y t
t ∗
E KK E
KKK
rrrr
τ K% r
y r π
M
Denote by DbR+ (kE ) the full triangulated subcategory of Db (kE ) consisting of
conic objects, that is, objects F such that for all j , H j (F ) is locally constant
on the orbits of the action of R+ on E. One defines the two functors
( • )∧
DbR+ (kE ) o / Db (k ∗ )
R+ E
( • )∨
Triangulated categories for the analysts 381
Example 4.2. (i) For a convex cone γ in E, denote Intγ its interior, by γ a = −γ
its image by the antipodal map and by γ ◦ its polar cone in E ∗ . Then, for a closed
convex proper cone γ with M ⊂ γ ,
(kγ )∧ kIntγ ◦ .
(ii) Let E denote the Euclidian space Rn and let x = (x1 , . . . , xn ) be the coordi-
nates. Let n = p + q with p, q ≥ 1 and set x = (x , x ) with x = (x1 , . . . , xp ),
x = (xp+1 , . . . , xn ). We denote by u = (u , u ) the dual coordinates on the
dual space E ∗ .
Let γ denote the closed solid cone in E,
γ = {x; x − x ≥ 0}
2 2
k∧γ kλ [−p].
H j (V ; νM (F )) lim H j (U ; F )
−→
U
382 Pierre Schapira
where U ranges over the family of open subsets of X which are “tangent” to V ,
that is, open tuboids in X with wedge M whose “profiles” is V . (For a precise
definition, refer to [16, § 4.2].)
The Sato’s microlocalization of F along M, denoted μM (F ), is the Fourier-
Sato transform of νM (F ), an object of DbR+ (kTM∗ X ). It satisfies:
Rπ ∗ μM (F ) RM (F ),
j
H j (μM (F ))(x0 ;ξ0 ) lim HU ∩Z (U ; F ),
−→
U,Z
Microdifferential operators
The sheaf of microfunctions allows us to analyze hyperfunctions microlocally,
that is, in the cotangent bundle. A similar localization may be performed with
respect to the sheaf of differential operators. Let X be a complex manifold of
dimension dX and denote by the diagonal of X × X. The sheaf of microlocal
operators is defined in [29] by
EXR := μ (OX×X
(0,dX )
) [dX ].
(0,dX )
Here, OX×X = OX×X ⊗q −1 OX q2−1 #X . One proves that EXR is concentrated in
2
degree 0 and is naturally endowed with a structure of a sheaf of C-algebras.
Moreover, for G ∈ Db (CX ), the object μhom (G, OX ) is well defined in Db (EXR )
(see [21]). This applies in particular to the sheaf CM of microfunctions.
The algebra EXR is extremely difficult to manipulate, but it contains the C-
algebra EX of microdifferential operators which is filtered and admits a symbol
calculus. When X is affine and U is open in T ∗ X, a section P ∈ ET ∗ X (U ) is
described by its “total symbol”
σtot (P )(x; ξ ) = pj (x; ξ ), m ∈ Z, pj ∈ (U ; OT ∗ X (j )),
−∞<j ≤m
However, this isomorphism is only locally defined and not unique. That is
why, in general, it is not possible to define such sheaves EX of microdifferential
operators on a homogeneous complex symplectic manifold X. However, Kashi-
wara [12] has shown that such a construction was possible when weakening
the notion of a sheaf of algebras by that of an algebroid stack. (Refer to [19]
for an introduction to stacks.)
and the total symbol of the product P ◦ Q is given by the Leibniz rule:
|α|
σtot (P ◦ Q) = ∂ α (σtot P )∂xα (σtot Q).
n
α! u
α∈N
Note that WT ∗ X has a natural filtration for which has order −1 and W
T ∗ X (0)
may naturally be considered as a deformation of the Poisson algebra OT ∗ X .
By replacing again the notion of a sheaf of algebras by that of an alge-
broid stack, one can define such objects on complex symplectic manifolds
(see [22, 27]).
A natural question would be to perform an analogous construction for
sheaves on real manifolds, that is, to construct a non conic microlocal the-
ory of sheaves on real symplectic manifolds. This problem is closely related to
Mirror Symmetry (see [25]).
Let us denote by
ρ: X −
→ Xsa (5.1)
Mod(kX ) o
ρ∗
/ Mod(k
Xsa ).
ρ −1
Definition 5.2. Let f ∈ CX∞ (U ). One says that f has polynomial growth at
p ∈ X if it satisfies the following condition. For a local coordinate system
(x1 , . . . , xn ) around p, there exist a sufficiently small compact neighborhood
K of p and a positive integer N such that
N
supx∈K∩U dist(x, K \ U ) |f (x)| < ∞ . (5.2)
temperate C ∞ -functions on Xsa . Note that the sheaf ρ∗ DX does not operate on
CX∞,t
sa
but ρ! DX does.
Now let X be a complex manifold. We still denote by X the real underlying
manifold and we denote by X the complex manifold conjugate to X. One
defines the sheaf of temperate holomorphic functions OX t
sa
as the Dolbeault
∞,t
complex with coefficients in CXsa . More precisely
OX
t
sa
= RHom ρ! DX (ρ! OX , CX∞,t
sa
). (5.3)
References
[1] E. Andronikof, Interview with Mikio Sato, Notices of the AMS, 54 Vol 2, (2007).
[2] I. Bernstein Modules over a ring of differential operators, Funct. Analysis and
Appl. 5 pp 89–101 (1971).
[3] A. D’Agnolo, Sheaves and D-modules in integral geometry, in Contemp. Math.,
251 Amer. Math. Soc., pp 141 (2000).
[4] A. D’Agnolo and P. Schapira, Leray’s quantization of projective duality, Duke
Math. Journ. 84 pp 453–496 (1996).
[5] O. Gabber, The integrability of the characteristic variety, Amer. Journ. Math. 103
pp 445–468 (1981).
[6] V. Guillemin, D. Quillen and S. Sternberg, The integrability of characteristics,
Comm. Pure and Appl. Math. 23 39–77 (1970)
[7] L. Hörmander, The analysis of linear partial differential operators, Grundlehren
der Math. Wiss. 256 Springer-Verlag (1983).
Triangulated categories for the analysts 387
Pierre Schapira
Université Pierre et Marie Curie
Institut de Mathématiques
175, rue du Chevaleret, 75013 Paris France
schapira@math.jussieu.fr
http://www.math.jussieu.fr/˜schapira/
Algebraic versus topological
triangulated categories
s tefan s ch w ede
These are extended and updated notes of a talk, the first version of which I gave
at the Workshop on Triangulated Categories at the University of Leeds, August
13–19, 2006. These notes are mostly expository and do not contain all proofs;
I intend to publish the remaining details elsewhere.
The most commonly known triangulated categories arise from chain com-
plexes in an abelian category by passing to chain homotopy classes or inverting
quasi-isomorphisms. Such examples are called ‘algebraic’ because they orig-
inate from abelian (or at least additive) categories. Stable homotopy theory
produces examples of triangulated categories by quite different means, and in
this context the source categories are usually very ‘non-additive’ before pass-
ing to homotopy classes of morphisms. Because of their origin I refer to these
examples as ‘topological triangulated categories’.
In this note I want to explain some systematic differences between these
two kinds of triangulated categories. There are certain properties – defined
entirely in terms of the triangulated structure – which hold in all algebraic
examples, but which fail in some topological ones. These differences are all
torsion phenomena, and rationally there is no difference between algebraic and
topological triangulated categories.
A triangulated category is algebraic in the sense of Keller [Ke, 3.6] if it is
triangle equivalent to the stable category of a Frobenius category, i.e., an exact
category with enough injectives and enough projectives in which injectives
and projectives coincide. Examples include all triangulated categories which
one should reasonably think of as ‘algebraic’: various homotopy categories
and derived categories of rings, schemes and abelian categories; stable module
categories of Frobenius rings; derived categories of modules over differential
graded algebras and differential graded categories. By a theorem of Porta [Po,
Thm. 1.2], every algebraic triangulated category which is well generated (a
mild restriction on its ‘size’, see [Ne, Def. 8.1.6 and 8.1.7]) is equivalent
389
390 Stefan Schwede
A short diagram chase shows that the group [X/n, X/n] is always annihilated
by n2 ; in algebraic triangulated categories, more is true:
⊗L : Db (Z) × T −→ T
X × [0, 1]
X = .
X × {0, 1} ∪ {x0 } × [0, 1]
Algebraic versus topological triangulated categories 391
here f
here g
f ∨g
X
pinch
/ X ∨ X / Y
392 Stefan Schwede
X × [0, 1] ∪X×{1} Y
C(f ) =
X × {0} ∪ {x0 } × [0, 1]
CX Y
the composite of any two is null-homotopic, and these sequences and their
(de-)suspensions give the distinguished triangles in SW.
is denoted by πns X and called the n-th stable homotopy group of X. For X = S 0 ,
we abbreviate πns S 0 to πns and speak about the n-th stable homotopy group of
spheres, also called the n-th stable stem. For example, π0s = Z, generated
by the identity map, and π1s = Z/2 generated by the class of the Hopf map
η : S 3 −→ S 2 . The stable stems are easy to define, but notoriously hard to
compute; for example, there is no finite CW-complex X which is non-trivial
in SW for which all stable homotopy groups are known! Much machinery of
algebraic topology has been developed to calculate such groups and understand
their structure, but no one expects to ever get explicit formulae for all stable
homotopy groups of spheres.
Algebraic versus topological triangulated categories 393
Under our assumption the first map is trivial so the smash product S/2 ∧ S/2
splits in SW as a sum of S/2 and S/2[1]. Thus as a module over the Steenrod
algebra, the mod-2 cohomology of the smash product S/2 ∧ S/2 decomposes
into a sum of two non-trivial summands.
On the other hand, there is a Künneth isomorphism for the mod-2 cohomol-
ogy of a smash product
H n (X ∧ Y, F2 ) ∼
= H p (X, F2 ) ⊗ H q (Y, F2 ) .
p+q=n
In the mod-2 cohomology of the mod-2 Moore spectrum the Bockstein oper-
ation Sq 1 : H 0 (S/2, F2 ) −→ H 1 (S/2, F2 ) is non-zero. The Cartan formula
shows that the operation Sq 2 is then non-trivial on the tensor product of
two copies of the generator of H 0 (S/2, F2 ). Thus the mod-2 cohomology
of S/2 ∧ S/2 is a 4-dimensional, indecomposable module over the Steenrod
algebra. We have reached a contradiction, which means that we must have
2 · S/2
= 0 in SW.
Algebraic versus topological triangulated categories 395
derived category D(A) of small differential graded category A [Po, Thm. 1.2].
The localization can be realized as a Bousfield localization of the ordinary
(i.e., ‘projective’) model structure on modules over A; hence D(A) and its
localizations are topological.
Examples of triangulated categories which are neither algebraic nor topo-
logical were recently constructed by Muro, Strickland and the author [MSS].
The simplest one is the category F(Z/4) of finitely generated free modules
over the ring Z/4. The category F(Z/4) has a unique triangulation with the
identity shift functor and such that the triangle
2 2 2
Z/4 −→ Z/4 −→ Z/4 −→ Z/4
is exact. For the proof of this and an argument why F(Z/4) is not topological
I refer to [MSS]. At present, I do not know any ‘exotic’ (i.e., non-topological
and non-algebraic) triangulated category in which 2 is invertible.
Unlike algebraic triangulated categories, we can not usually expect that a
topological triangulated category can be tensored over Db (Z), the bounded
derived category of finitely generated abelian groups. The appropriate replace-
ment for Db (Z) is the Spanier-Whitehead category: for every topological trian-
gulated category T , there is a biexact pairing
∧ : SW × T −→ T
∧ : HoSSet∗ × Ho C −→ Ho C ,
which uses a technique called ‘framings’, see [Ho1, Thm. 5.7.3]. The geometric
realization of a finite simplicial set is a finite CW-complex, and up to homo-
topy equivalence, every finite CW-complex arises in this way. Since moreover
suspension is invertible in Ho C, this extends to a well-defined pairing on the
Spanier-Whitehead category.
There is an alternative construction for spectral model categories, i.e., model
categories C which are enriched over the category Sp of symmetric spectra
of [HSS], compatibly with the smash product and the stable model structure
(see [SS, Def. 3.5.1] for details). The derived smash product between symmetric
Algebraic versus topological triangulated categories 397
vanishes and so the group [S/n, S/n] is cyclic of order n for n odd, hence
n · S/n = 0. (For n = 2 we have π1 (S/2) ∼= Z/2 and the analogous short exact
sequence does not split by Proposition 4. So [S/2, S/2] ∼
= Z/4.)
The rest of the argument is then the same as in Proposition 1. For an odd
prime p we can smash the distinguished triangle
p·
S −→ S −→ S/p −→ S[1]
From what we have discussed so far, it is still conceivable that every topo-
logical triangulated category in which 2 is invertible is algebraic. In particular
one can wonder whether the Spanier-Whitehead category is algebraic after
localization at an odd prime. We now describe a property of triangulated cate-
gories which distinguishes topological from algebraic examples away from the
prime 2.
As before we denote by K/n any cone of n · IdK , which comes as part of a
distinguished triangle
n· π
K −
→ K −→ K/n −→ K[1] .
where μ : Z/n ⊗L Z/n −→ Z/n is the multiplication map which makes Z/n
into a ring. We choose a distinguished triangle
f¯ ϕ δ
Z/n ⊗L K −→ Z/n ⊗L X −→ C(f¯) −
→ Z/n ⊗L K[1] (8)
Z/n ⊗L Z/n ⊗L K
Z/n⊗f¯
/ Z/n ⊗L Z/n ⊗L X Z/n⊗ϕ
/ Z/n ⊗L C(f¯) Z/n⊗δ
/ Z/n ⊗L Z/n ⊗L K[1]
μ⊗K μ⊗X σ μ⊗K[1]
Z/n ⊗L K / Z/n ⊗L X / C(f¯) / Z/n ⊗L K[1]
f¯ ϕ δ
whose lower row is the distinguished triangle (8) and whose upper row is (8)
tensored from the left with Z/n. The left square commutes since the multipli-
cation morphism μ is associative in Db (Z). Since both rows are distinguished
triangles, there exists a morphism σ : Z/n ⊗L C(f¯) −→ C(f¯) making the
middle and right square commute.
We consider the morphism
π ⊗ C(f¯) : C(f¯) ∼
= Z ⊗L C(f¯) −→ Z/n ⊗L C(f¯)
β̄ : S/3[11] −→ S/3
with 3 · C(β̄) = 0.
Since the stable stems in dimension 21 and 22 consist only of torsion
prime to 3 we have π22 (S/3) = [S[22], S/3] = 0. So there is a morphism
a : S[23] −→ C(β̄) lifting β̃1 in the sense that δa = β̃1 [12]. Since we assumed
3 · C(β̄) = 0, the morphism a can be extended to a morphism ā : S/3[23] −→
Algebraic versus topological triangulated categories 401
Since the stable stems in dimension 21, 22, 33 and 34 consist only of torsion
prime to 3 we have π34 C(β̄) = [S[34], C(β̄)] = 0 and so there is a morphism
b : S[35] −→ C(ā) lifting β̃1 in the sense that δb = β̃1 [24].
Now we bring cohomology operations into the game to reach a contra-
diction. The Moore spectrum S/3 has its mod-3 cohomology concentrated in
dimensions 0 and 1, where it is 1-dimensional. The mapping cone of b is built
by distinguished triangles from three shifted copies of S/3 and one shifted copy
of S, so its mod-3 cohomology is concentrated in dimensions 0, 1, 12, 13, 24,
25 and 36, where it is 1-dimensional.
The morphism β̃1 is detected by the Steenrod operation P 3 , i.e., in the mod-3
cohomology of the mapping cone of β̃1 the operation P 3 is non-trivial from
dimension 0 to dimension 12. The stable ‘cells’ (i.e., shifted copies of S) of
the mapping cone of b in dimension 12, 24 and 36, are attached to the two
cells directly below by β̃1 , so in the cohomology of the mapping cone of b,
the 3-fold iterate of P 3 is non-trivial from dimension 0 to 36. By the Adem
relations we have (P 3 )3 = (P 7 P 1 − P 8 )P 1 . Since the cohomology is trivial
in dimension 4 the operation P 1 acts trivially, hence so does (P 3 )3 . We have
obtained a contradiction, and so no extension of β̃1 has a mapping cone which
is annihilated by 3.
Proof. The key point is that if n is prime to 6, then the mod-n Moore spectrum
has an associative multiplication in the Spanier-Whitehead category (the mul-
tiplication is also commutative, but that is not relevant for the current proof).
Again I cannot refrain from giving the simple proof below. In contrast, the mod-
2 Moore spectrum does not have a multiplication; the mod-3 Moore spectrum
has a commutative multiplication, but that is not associative in SW, see [To].
As in the proof of Proposition 5, the condition that n is odd guarantees that
n · S/n = 0. So there exists a morphism μ : S/n ∧ S/n −→ S/n which splits
402 Stefan Schwede
factors as a composite
δ∧δ∧δ
S/n ∧ S/n ∧ S/n −−−→ S[3] −→ S/n
where δ : S/n −→ S[1] is the connecting morphism. We have π3s ∼ = Z/24 and
π2s ∼
= Z/2, so if n is prime to 6 we have [S[3], S/n] = π3 (S/n) = 0. Thus the
associator is trivial or, equivalently, μ is associative in SW.
The rest of the argument is now essentially the same as in Proposition 7. We
can smash the distinguished triangle
n· δ
S −→ S −→ S/n −
→ S[1]
Then we use the same reasoning as in the proof of Proposition 7 to obtain the
retraction σ : S/n ∧ C(f¯) −→ C(f¯) which shows that n · C(f¯) = 0.
Remark 14. The proof of Proposition 12 shows that the special features of topo-
logical over algebraic triangulated categories are closely related to existence
and properties of multiplications on mod-n Moore spectra. For primes p ≥ 5,
the mod-p Moore spectrum has a multiplication in the Spanier-Whitehead
category which is commutative and associative. So on the level of tensor trian-
gulated categories, there does not seem to be any qualitative difference between
the Moore spectrum S/p as an object of the Spanier-Whitehead category SW
and Z/p as an object of Db (Z), as long as p ≥ 5. However, mod-n Moore
spectra are never A∞ ring spectra, but rigorously defining what that means
and proving it would lead us too far afield. Theorem 17 below explains how
the higher order non-associativity eventually manifests itself in the triangulated
structure of the Spanier-Whitehead category (i.e., without any reference to the
smash product).
Algebraic versus topological triangulated categories 403
From what we have discussed so far, it is still conceivable that for primes
p ≥ 5 the p-local Spanier-Whitehead category is algebraic. We will now intro-
duce an invariant which we then use to show that this is not the case for any
prime p.
The last two items show that the n-order is useless if T is a k-linear triangu-
lated category for some field k, since then every n ∈ Z is either zero or a unit
in k.
The results which we have obtained so far can be rephrased using the notion
of n-order: if T is algebraic, then Propositions 1 and 7 show that for every
object X, the object X/n always has n-order at least 2; we will improve this in
Theorem 16 below. Propositions 4 and 11 show that in the Spanier-Whitehead
category, the Moore spectrum S/2 has 2-order 0 and S/3 has 3-order 1; we
generalize this to mod-p Moore spectra in Theorem 17 below. If T is topological
then for every object X, the object X/3 has 3-order at least 1, by Proposition 5.
If T is topological and p is a prime ≥ 5, then for every object X, the object
X/p has p-order at least 2, by Proposition 12.
The following theorem generalizes Propositions 1 and 7.
a distinguished triangle
fˆ
ρ∗ K −→ Z −→ C(fˆ) −→ ρ∗ K[1]
The next theorem generalizes Propositions 4 and 11 and shows that topo-
logical triangulated categories behave quite differently from algebraic ones.
The proof of Theorem 17 has two parts. One ingredient is a general state-
ment about p-orders in topological triangulated categories T which generalizes
Propositions 5 and 12: for any object X and prime p, the object X/p has p-
order greater or equal to p − 2. The proof of this result uses the concept and
properties of a coherent action of a mod-p Moore space on an object of a model
category, see Section 2 of [Sch]. It follows that any topological triangulated
category (such as the Spanier-Whitehead category) has p-order at least p − 1.
The second ingredient of Theorem 17 is the proof that the mod-p Moore
spectrum S/p has p-order at most p − 2. This uses mod-p cohomology oper-
ations and serious calculational input; in particular, the proof depends on van-
ishing results, due to other people, about the p-primary components of the
stable stems in specific dimensions. Proposition 11 gives the flavor of the proof
which, however, for general primes p is more involved. I plan to give a detailed
proof elsewhere.
As we just mentioned, every topological triangulated category has p-order
at least p − 1. This leave us with the following question, which generalizes
Problems 6 and 13
Open problem 18. Let p be a an odd prime. Does there exist a triangulated
category whose p-order is strictly less than p − 1?
More generally we can ask which values the n-orders of triangulated cate-
gories can take.
Now that we have discussed torsion phenomena which can distinguish alge-
braic from topological triangulated categories, it is natural to ask whether there
are any differences between topological and algebraic triangulated categories
if all primes are invertible, i.e., in Q-linear triangulated categories. The n-order
406 Stefan Schwede
References
[Ad] J. F. Adams, Stable homotopy and generalised homology. Chicago Lectures
in Mathematics. University of Chicago Press, Chicago, Ill.-London, 1974.
x+373 pp.
[BF] A. K. Bousfield, E. M. Friedlander, Homotopy theory of -spaces, spectra,
and bisimplicial sets. Geometric applications of homotopy theory (Proc.
Conf., Evanston, Ill., 1977), II Lecture Notes in Math., vol. 658, Springer,
Berlin, 1978, pp. 80–130.
[Du] D. Dugger, Spectral enrichments of model categories. Homology, Homotopy
and Applications 8 (2006), 1–30.
[EKMM] A. D. Elmendorf, I. Kriz, M. A. Mandell, J. P. May, Rings, modules, and alge-
bras in stable homotopy theory. With an appendix by M. Cole, Mathematical
Surveys and Monographs, 47, American Mathematical Society, Providence,
RI, 1997, xii+249 pp.
[He] A. Heider, Two results from Morita theory of stable model categories.
!%!%!%!%
Algebraic versus topological triangulated categories 407
Derived categories of coherent sheaves on
algebraic varieties
yukinobu toda
Abstract. In this article, we give the introduction of the recent developments
on the derived categories of coherent sheaves on algebraic varieties. We also
introduce the notion of stability conditions on triangulated categories in the sense
of T. Bridgeland.
1. Introduction
The notion of derived category of coherent sheaves was first introduced in [24]
in order to formulate the Grothendieck duality theorem. It is a category whose
objects are complexes of coherent sheaves, and has a structure of a triangulated
category. Recently it has been observed that the derived category represents
several interesting symmetries, which seems impossible without the notion
of derived categories, e.g. McKay correspondence [16], Homological mirror
symmetry [43], etc. Now derived categories are a very popular area with interac-
tions with many other subjects including non-commutative algebra, birational
geometry, symplectic geometry and string theory. In this article, we give an
introduction of the recent results on these topics.
The content of this article is as follows. In Section 2, we give the basic
notions concerning derived categories, and propose some fundamental prob-
lems. In Section 3, we discuss the relationship between derived category
and birational geometry. In Section 4, we discuss the symmetries between
derived categories of coherent sheaves and that of module categories of
some non-commutative algebras. In Section 5, we introduce the notion of
stability conditions on triangulated categories, defined by T. Bridgeland,
and see how this notion explains the several symmetries we discuss in this
article.
Notation and Convention. All the varieties are defined over C. For a variety
X, we denote by Coh(X) the abelian category of coherent sheaves on X. When
X is smooth, we denote by ωX := ∧dim X #X the canonical line bundle, and
KX is the divisor on X such that OX (KX ) = ωX . For a C-vector space V , we
denote by V ∗ its dual vector space.
408
Derived categories of coherent sheaves on algebraic varieties 409
D(X) := D b Coh(X),
· · · → 0 → F m → F m+1 → · · · → F n−1 → F n → 0 → · · · ,
H• C
{{ CC g
f {
{ CC
{{ CC
}{{ C!
F• G•,
Also these notions should satisfy some axioms. (cf. [24].) On the category
D(X), the shift functor [1] is given by shifting the degree of the complex, and
distinguished triangles are obtained by taking mapping cones of the morphisms
of chain complexes.
Note that the abelian category Coh(X) is regarded as the full subcategory of
D(X), by identifying the objects in Coh(X) with the complexes concentrated
in degree zero. For an object E ∈ D(X), we denote by Hi (E) ∈ Coh(X) the
i-th cohomology of the complex E. The support of E ∈ D(X) is defined by
Supp(E) = Supp Hi (E) ⊂ X.
i
Here we remark that if E and F are contained in Coh(X), then Homi (E, F )
coincides with the usual Ext-group Exti (E, F ). If X is smooth and projective,
410 Yukinobu Toda
Now we introduce some functors between derived categories. For the precise
definitions and details, see [24]. Suppose X is smooth and Y is another smooth
variety. Let f : X → Y be a projective morphism. Then we can define the
derived push-forward and derived pull-back,
The functor Rf∗ is defined without the assumption that Y is smooth. However
Lf ∗ is not defined if Y is singular. In fact Lf ∗ F for F ∈ D(Y ) may be
unbounded below, hence one has to replace D(X) by the unbounded (or bounded
above) derived category of coherent sheaves. Also for an object E ∈ D(X), we
can define the derived homomorphism and derived tensor product,
L
RHom(E, ∗) : D(X) → D(X), ∗ ⊗ E : D(X) → D(X).
L
P ∗
X→Y (E) = RpY ∗ (pX E ⊗ P),
Example 2.3. (Mukai [48], [49]) (i) Let A be an abelian variety, i.e. A is
a complex torus Cn / embedded into a projective space. Here ⊂ R2n is a
lattice of rank 2n. Let  = Pic0 (A) be the moduli space of line bundles on
A which are topologically isomorphic to OA . It is known that  is also an
abelian variety of the same dimension, and called dual abelian variety. Let
U ∈ Pic(A × Â) be the universal family. Then the functor
U
Â→A
: D(Â) −→ D(A),
is a Fourier-Mukai transform. See [48] for the detail. When dim A ≥ 2, the
dual  is not necessary isomorphic to A. This means that the derived category
D(A) does not determine the original variety A.
(ii) An analogue of (i) for K3 surfaces was studied in [48]. Let X be a K3
surface, i.e. simply connected algebraic surface with ωX trivial. Let M be a
connected component of the moduli space of stable sheaves on X, which is two
dimensional, projective, and has a universal family E ∈ Coh(X × M). Then the
functor
E
M→X : D(M) −→ D(X),
is a Fourier-Mukai transform.
412 Yukinobu Toda
: D(Y ) −→ D(X),
for E, F ∈ T .
Definition 2.7. [8, Definition 2.1] An object P ∈ D(X) is called a point object
of codimension s if it satisfies,
(i) SX (P ) ∼
= P [s],
(ii) Hom<0 (P , P ) = 0,
(iii) Hom0 (P , P ) = C.
Now the idea of the proof of Theorem 2.5 is explained as follows. Note
that the notion of point objects are determined by only the data of D(X) as a
triangulated category. In [8, Proposition 2.2], it is shown that if KX or −KX
is ample, then any point objects are of the form Ox [r] for some closed point
x ∈ X and r ∈ Z. Thus one can reconstruct the notion of “point” from the
data of D(X) as a triangulated category, which enables us to reconstruct X
from D(X).
On the other hand if KX is trivial, then the condition (i) of Definition 2.7 does
not say anything so there may be other point objects, and non-trivial Fourier-
Mukai partners. In fact the examples given in Example 2.3 are such cases.
At this time, very few is known about Fourier-Mukai partners. Other than
Theorem 2.5, we know the following cases:
Â × A −→ B̂ × B.
D(X) / D(Y )
√ √
ch(∗) tdX ch(∗) tdY
H ∗ (X, Q)
φ
/ H ∗ (Y, Q).
In particular we have κ(X) = κ(Y ). (Also see [37].) Here κ(X) is the Kodaira
dimension, defined by the number k ∈ Z such that
H 0 (X, mKX ) ∼ amk , (m 0) (1)
for some a > 0 if H 0 (X, mKX ) is non-zero for some m > 0. If H 0 (X, mKX ) =
0 for any m > 0, κ(X) is defined to be −∞. The idea of [60] is to reduce the
classification problem of Fourier-Mukai partners to the case of Kodaira dimen-
sion zero, by comparing the derived categories of the fibers of the following
rational maps called Iitaka fibrations,
X Proj H 0 (X, mKX ), Y Proj H 0 (X, mKY ).
m≥0 m≥0
The above space parameterizes the first order deformations of the abelian cate-
gory Coh(X). (See [45], [46], [64].) In fact for u = (α, β, γ ), β corresponds to
the usual deformation of X as a scheme, γ corresponds to a non-commutative
deformation, and α is a gerby deformation. One can construct the C[ε]/(ε 2 )-
linear abelian category Coh(X, u), which is a mixture of the above deforma-
tions. Then the equivalence extends to an equivalence [64],
†
: D b Coh(X, u) −→ D b Coh(Y, v),
When KX or −KX is ample, Bondal and Orlov [8] showed (by the same idea
of Theorem 2.5) that the group Auteq D(X) is generated by trivial autoequiva-
lences, that is the group of automorphisms Aut(X), tensoring line bundles, and
the shift functor [1]. Again it seems that Auteq D(X) is more difficult when
the Kodaira dimension of X is near to zero. Non-trivial autoequivalences are
provided by the spherical objects below.
R Hom(E, F ) ⊗ E −→ F −→ TE (F ).
and define f: A × Â → A × Â to be
δ̂ −γ̂
f = .
−β̂ α̂
Theorem 2.11. (Orlov [53]) For an abelian variety A, there exists the following
exact sequence,
Z = C1 ∪ · · · ∪ Cn , Ci ∼
= P1 .
418 Yukinobu Toda
Under this situation, Ishii and Uehara [29] considered the triangulated category,
Note that OCi (−1), ωZ are spherical objects, so there are associated twists
TOCi (−1) , TωZ . We set
Furthermore Ishii, Ueda and Uehara [28] determined the group structure of
B using the space of stability conditions.
such that each Xi Xi+1 is a birational morphism which contracts a divisor,
or a certain birational map called flip. The output Xm is either a minimal model
(i.e. KXm · C ≥ 0 for any curve C ⊂ Xm ) or has a fiber space structure Xm → Z
such that the general fiber is a Fano manifold. The purpose of this section is to
observe that MMP is more natural in the context of the “space” stated in the
introduction.
One can show the following: for a full subcategory C ⊂ T , it is right admis-
sible if and only if for E ∈ T , there is a distinguished triangle,
E1 −→ E −→ E2 ,
T =< C1 , C2 , · · · , Cm >,
E
i /Y
p f
Z / X.
Let OE (1) ∈ Pic(E) be the tautological line bundle of the projective bundle
E = P(NZ/X ). Bondal and Orlov [7] showed the following.
Theorem 3.3 asserts that the derived category will be bigger by taking blow-
ups, and the difference is described by the derived category of the exceptional
locus.
Definition 3.4. Let X and X† be smooth quasi projective varieties and assume
there exist projective birational morphisms,
f g
X −→ Y ←− X† ,
Remark 3.5. Usually flips and flops are defined over varieties with terminal
singularities or pairs (X, B) with klt singularities, because MMP works in the
category of varieties or pairs with such singularities. (See [44].) However we
omit this general treatment in order to simplify the argument.
In [7], Bondal and Orlov studied the behavior of the derived categories
under certain special flips and flops. Let f : X → Y be as in Definition 3.4
with dim X = n. Suppose the exceptional locus E is isomorphic to Pr with
normal bundle OE (−1)⊕n−r . Then we have the diagram,
ZA
~~ AA
p
~~ AAq
~~ AA
~
~ A
X? X†
?? }}
??
? }}}g
f ??
~}
}
Y,
It is known that any birational minimal models are connected by flops [35].
Hence Theorem 3.7 leads us to the following conjecture.
Theorem 3.3, Theorem 3.7 and Conjecture 3.8 indicate the following. One of
the features of MMP in dimension bigger or equal to three is that the minimal
model is not necessary unique. However if Conjecture 3.8 is true, then it is
unique on the level of derived categories. Hence if we have some kind of nice
geometry of “spaces” as in the introduction, and an appropriate theory of the
analogue of minimal model theory, then the corresponding minimal model must
be unique. Furthermore, because derived categories will be smaller by blow
downs and flips by Theorem 3.3, Theorem 3.7, the minimal model theory in
the “space” should correspond to “minimizing” derived categories. This is the
reason why the geometry needed for string theory is also natural in birational
geometry, and hopefully it will enable string theory to interact with birational
geometry. However there are several technical problems in realizing this story,
and it is only just a philosophy at this time.
422 Yukinobu Toda
τ≤0 E −→ E −→ τ≥1 E,
The subcategory
T ≤0 ∩ T ≥1 [1] ⊂ T ,
Proposition 3.11. ([11]) There are t-structures on D(X) with hearts i Per(X/Y )
for i = −1, 0 such that an object E ∈ D(X) is contained in i Per(X/Y ) if and
only if
r Hj (E) = 0 unless j = −1, 0.
r R 1 f∗ H0 (E) = 0 and f∗ H−1 (E) = 0.
r For any c ∈ Coh(X) with Rf∗ c = 0, we have
Z@
@@ q
p @@
@@
X Y,
such that p, q are birational and p ∗ KX = q ∗ KY . It is well known that if X and Y
are birational minimal models, then they are K-equivalent. Also Kawamata [37]
called the two varieties X, Y D-equivalent if Y is a Fourier-Mukai partner of X.
As a generalization of Conjecture 3.8, Kawamata [37] proposed the following
conjecture.
Conjecture 3.13. (Kawamata [37]) For birationally equivalent smooth pro-
jective varieties X and Y , they are K-equivalent if and only if they are D-
equivalent.
Uehara [66] found an example of birationally equivalent varieties X and
Y whose derived categories are equivalent, but they are not K-equivalent.
Thus Conjecture 3.14 is not true in the sense that D-equivalence does not
imply K-equivalence. However there are some results, including Theorem 3.7,
Theorem 3.12, which support the conjecture that K-equivalence implies D-
equivalence. Here we put another evidence for this.
Theorem 3.14. (Kawamata [37], Namikawa [50]) Suppose that X and X †
are related by a Mukai flop X → Y ← X† , i.e. there is a diagram
ZA
AA p
p1
AA 2
AA
X? X†
?? ~
?? ~~
?
f1 ?? ~~~f2
~~
~
Y ,
Here mod A is the abelian category of right A-modules, and the functor 2 is
forgetting the A-module structure via the natural morphism R → A. By abuse
of notation, we also use the notation R Hom(E, ∗) for the functor 1 . As an
analogue of Morita correspondence, we have the following theorem.
One can consult [25, Theorem 7.6] for the proof of Theorem 4.2.
T =< E1 , · · · , En > .
n
E= Ei ,
i=1
where A = End(E). This sort of equivalence was first studied by Beilinson [2].
Theorem 4.4. (Beilinson [2]) We have the following description of the derived
category of Pn .
D(X) ∼
= DG (Cn ).
Theorem 4.7. (Bridgeland, King, Reid [16]) Conjecture 4.6 is true when
n ≤ 3.
Proof. We explain the idea of [16], which is similar to the proof of Theo-
rem 3.12. First we consider the G-clusters on C3 . These are G-invariant closed
subschemes Z ⊂ C3 which satisfy OZ ∼ = C[G]. Then consider the moduli
space of G-clusters, denoted by G- Hilb(C3 ). By associating a G-cluster to its
support, one has the map
G- Hilb(C3 ) −→ C3 /G.
Z
x GGG
q xxx GGp
xx GG
x GG
|x
x #
XD
DD C3
DD yy
DD yy
f DD yyyπ
" |y
C3 /G.
gives an equivalence.
428 Yukinobu Toda
Their method is very different from the proof of Theorem 4.7, using charac-
teristic p methods and deformation quantization. Another special case is due
to Kawamata [40].
Definition 4.10. (M. Van den Bergh [20]) A non-commutative crepant resolu-
tion of R is an homologically homogeneous R-algebra of the form A = End(M)
where M is a reflective R-module.
Theorem 4.11. (M. Van den Bergh [21]) Assume dim f −1 (p) ≤ 1 for any
p ∈ Spec R. Then there exists a tilting generator E ∈ D(X) such that A =
End(E) is a non-commutative crepant resolution. In particular there is a derived
Derived categories of coherent sheaves on algebraic varieties 429
equivalence,
i: D(X) −→ D(mod Ai ),
for i = −1, 0, where Ai = End(Ei ). The algebras A−1 , A0 are path algebras
with certain relations corresponding to the quiver given in Figure 4.13. Espe-
cially the category mod Ai has two simple objects, corresponding to the two
vertex in the quiver. In [21], it is shown that i takes i Per(X/Y ) to mod Ai .
The corresponding simple objects in −1 Per(X/Y ) are given by
respectively.
Figure 4.13.
(
•h (•
h
In dimension three, M. Van den Bergh [20] also proved the following.
Theorem 4.14. (M. Van den Bergh [20]) Assume that R is three dimen-
sional and has terminal singularities. Then R has a non-commutative crepant
resolution if and only if it has a commutative one. Furthermore any two
crepant resolutions (commutative as well as non-commutative ones) are derived
equivalent.
430 Yukinobu Toda
A1 A2 An
A1 A2 An
Here Z is called the central charge. Each P(φ) is an abelian category, the
non-zero objects of P(φ) are called semistable of phase φ, and the non-zero
simple objects of P(φ) are called stable. The objects Aj are called semistable
factors of E with respect to σ . We write φσ+ (E) = φ1 and φσ− (E) = φn . The
mass of E is defined to be
n
mσ (E) = |Z(Ai )|.
i=1
(Z, P)
−→ (Z, P((0, 1])),
Remark 5.4. In general, the category P((a, b)) for b − a < 1 is not an abelian
category, and it is only a quasi-abelian category. (See [14, Section 4].)
Now we give some examples using Proposition 5.3.
Example 5.5. (i) Let C be a smooth projective curve and D = D(C). Then
setting A = Coh(C) and Z : K(C) → C by
Z(E) = − deg(E) + i rank(E),
gives a stability condition on D. In the language of Definition 5.2, P(φ) ⊂ D
for 0 < φ ≤ 1 is given by
P(φ) = {semistable sheaves E on C with μ(E) = −1/ tan(π φ).}
(ii) For a field k, let A be a finite dimensional k-algebra. Let modf (A)
be the abelian category of finite dimensional right A-modules, and D the
bounded derived category of modf (A). The abelian category modf (A) has
finite number of simple objects, say S1 , · · · , SN . Then setting A = modf (A)
and Z : K(D) → C arbitrary so that Im Z(Si ) > 0, (it is possible because the
classes [S1 ], · · · , [SN ] in K(D) form a basis of K(D)), gives a stability condi-
tion on D.
K(X)
Z / C.
uu:
u
uu
ch
uuu
u
H ∗ (X, Q)
We write Stab(X) for the set of locally finite numerical stability conditions on
D.
Let K0 (X) be the kernel of the map ch : K(X) → H ∗ (X, Q), and set N (X)
as follows:
N (X) = K(X)/K0 (X).
The group N (X) is called numerical Grothendieck group. Note that we have
the following injection,
1
ch : N (X) −→ H i (X, Z),
i!
hence in particular N (X) is a finitely generated Z-module. We have the fol-
lowing map,
Z : Stab(X) −→ Hom(N (X), C).
Because the dimension of Hom(N (X), C) is finite, Theorem 5.6 implies that
any connected component of Stab(X) is a finite dimensional complex manifold.
Lemma 5.9. For a triangulated category D, the space Stab(D) carries the
: + (2, R), and the left action of the group Auteq(D).
right action of the group GL
+
: (2, R), define
Proof. Given a stability condition σ = (Z, P) and (T , f ) ∈ GL
the stability condition σ = (Z , P ) by setting
Z = T −1 ◦ Z, P (φ) = P(f (φ)).
: + (2, R) on Stab(D).
Then it gives the right action of GL
Next for σ = (Z, P) ∈ Stab(D) and ∈ Auteq(D), its left action is defined
by (σ ) = (Z , P ) where
Z = Z ◦ −1
, P (φ) = (P(φ)).
+
: (2, R) and the restriction of the action
Note that there is a subgroup C ⊂ GL
: + (2, R) to C is free. Explicitly for λ ∈ C, its action on σ = (Z, P) is
of GL
described by σ = (Z , P ) where
Z = e−iπλ Z, P (φ) = P(φ + Re(λ)).
subset UX ⊂ M which comes from sigma model constructions from the triples
(6) is an open subset. We call UX the neighborhood of the large volume limit,
and let M(X) be the connected component of M which contains UX . Here
we note that there might be some points in M(X) which correspond to sigma
models for topologically distinct Calabi-Yau 3-folds, or do not correspond to
sigma models at all.
There are two foliations on M, one of them is when restricted to UX , fixing
the complex structure I and deforming β + iω. Another foliation is fixing
β + iω and deforming I on UX . Let us consider the leaves of the corresponding
foliations,
The space MC (X) is nothing but the moduli space of the complex structures,
which is defined mathematically and has been studied in algebraic geom-
etry up to now. However the space MK (X), called stringy Kähler moduli
space, is not defined mathematically rigorous way, as well as sigma model
constructions.
Now let us discuss on mirror symmetry. There is an involution m : M → M
called the mirror map, which exchanges the above two foliations on M. Then
a triple
is called a mirror of the triple (6) if the associated sigma model is mapped to the
sigma model of (6) by the mirror map m. Especially we have the isomorphisms,
MK (X) ∼
= MC (X̂), MK (X̂) ∼
= MC (X).
D b Fuk(X, β + iω) ∼
= D b Coh(X̂, Iˆ), D b Coh(X, I ) ∼
= D b Fuk(X̂, β̂ + i ω̂),
for a mirror (X̂, β̂ + i ω̂, Iˆ). This is the Homological mirror symmetry proposed
by Kontsevich [45].
The fact that D b Coh(X, I ) does not depend on β + iω imply that the associ-
ated categories of B-models are unchanged along the subspace MK (X) ⊂ M.
Below we fix the complex structure I and simply write X for the complex man-
ifold (X, I ). The important point is that the category of particular B-branes,
called BPS-branes, are changed along MK (X) in the fixed category of B-
branes D(X). Let us take a point p ∈ MK (X) and the associated category of
BPS branes B(p). Here we remark that the point p does not have the informa-
tion how the category B(p) is embedded into D(X). Now let us take a closed
path,
γ : [0, 1] −→ MK (X),
with γ (0) = γ (1) = p, and choose an embedding B(p) ⊂ D(X). Then for each
t ∈ [0, 1], we have the associated subcategory,
For t = 1, we obtain the embedding of B(γ (1)) ⊂ D(X), and B(γ (1)) = B(p).
So we obtain the two embeddings of B(p) into D(X) for t = 0, 1, which
may be different. In fact its difference should be induced by the action of an
autoequivalence on D(X), hence we have the map,
Now recall that the space Stab(X) carries a left action of Auteq D(X). Also
it is expected that for σ = (Z, P) ∈ Stab(X), the associated category
P(φ),
φ∈R
defines the category of BPS branes at some point of MK (X). Combined these
observations, it is reasonable to guess that the space Auteq D(X)\ Stab(X) is
related to the space MK (X). More precisely Bridgeland conjectures in [9] that
Derived categories of coherent sheaves on algebraic varieties 437
Note that the RHS of (8) is nothing but the moduli space of elliptic curves.
Hence in this case, we have obtained the complete picture.
Under the identification (10), the bilinear pairing −χ (E1 , E2 ) on the left hand
side goes to the pairing (9). Let us define Z(β,ω) : N (X) → C by the formula.
Z(β,ω) (E) = (exp(β + iω), v(E)). (11)
Suppose v(E) = (r, l, s) with r
= 0. Then (11) is written as
1 2
Z(β,ω) (E) = (l − 2rs) + r 2 ω2 − (l − rβ)2 + i(l − rβ) · ω. (12)
2r
If r = 0, (11) is written as Z(E) = (−s + l · β) + i(l · ω). We define σ(β,ω) to
be the pair (Z(β,ω) , A(β,ω) ).
Proposition 5.13. [15, Lemma 6.2, Proposition 7.1] The subcategory
A(β,ω) ⊂ D(X) is a heart of a bounded t-structure, and the pair σ(β,ω) gives a
stability condition on D(X) if and only if for any spherical sheaf E on X, one
has Z(β,ω) (E) ∈
/ R≤0 . This holds whenever ω2 > 2.
Let Stab∗ (X) be the connected component of Stab(X) which contains σ(β,ω) .
Note that the pairing (9) and the isomorphism (10) induces an isomorphism,
Hom(N (X), C) ∼
= NS∗ (X).
Hence we have the map,
Z : Stab∗ (X) −→ NS∗ (X).
In order to understand the image of Z, first define the subset
P(X) ⊂ NS∗ (X)C ,
to be the set of v = v1 + iv2 ∈ NS∗ (X)C with vi ∈ NS∗ (X) such that v1 , v2
span a positive definite two plane in NS∗ (X). The space P(X) consists of
two connected components, and let P + (X) ⊂ P(X) be the component which
contains (1, iω, − 12 ω2 ). We define P0+ (X) to be
P0+ (X) = P0 (X) \ δ⊥.
δ∈(X)
∗
Here (X) = {v ∈ NS (X) | (v, v) = −2}.
Theorem 5.14. (Bridgeland [15, Theorem 1.1]) The component Stab∗ (X) is
mapped by Z onto the open subset P0+ (X), and the induced map
Z : Stab∗ (X) −→ P0+ (X),
is a regular covering map. Its Galois group G fits into the exact sequence,
0 −→ G −→ Auteq∗ D(X) −→ Aut H ∗ (X, Z).
440 Yukinobu Toda
assuming the volumes of all the curves in X except C are big enough. The
resulting picture is
MK (X) = P1 \ {0, 1, ∞} (13)
†
= U K (X) ∪ U K (X ), (14)
and UK (X) ∩ UK (X† ) = ∅. Here string theory has singularities at one of the
deleted points {0, 1, ∞}, and the other two points are large volume limits
corresponding to X and X† .
Let us translate the above argument to the space Stab(X). First we can
guess that there exists some open subset UX ⊂ Stab(X), which corresponds to
the neighborhood of the large volume limit. For any point σ ∈ UX , the corre-
sponding t-structures given in Proposition 5.3 should be related to the standard
t-structure, for example obtained by tilting. Next let W and X be three dimen-
sional birational minimal models, and : D(W ) → D(X) an equivalence of
derived categories. Then induces an homeomorphism,
such that two chambers are either equal or disjoint. This picture is quite similar
to the movable cone [36, Theorem 2.3]. Thus by describing the space Stab(X),
we can observe the relationship between SCFT and birational geometry.
Zn Z
N 1 (X/Y )C / K(X/Y )∗C .
Stabn (X† /Y )
∗
/ Stabn (X/Y )
N 1 (X† /Y )C
φ∗
/ N 1 (X/Y )C .
Here φ : X† X is the birational map. By Theorem 3.12 and Example 4.12,
the equivalence restricts to the equivalence,
0
Per(X† /Y ) ⊗ OX† (1) ∩ D(X/Y ) −→ 0 Per(X/Y ) ∩ D(X/Y ).
This means ∗ induces a homeomorphism,
f : H ev (M, C) −→ C,
Stab(A) = {(Z, P) ∈ Stab(D) | P((0, 1]) = A and Im Z([Si ]) > 0 for all i}.
Example 5.19. (i) Suppose for E, F ∈ A one has dim Exti (E, F ) = 0 for
i ≥ 2. Then
satisfies (17).
446 Yukinobu Toda
with
f α : Stab(A) −→ C,
using counting invariants of semistable objects and satisfies certain p.d.e. Now
we give the outline of the construction of f α in [33]. For each α ∈ C(A) and
σ ∈ Stab(A), Joyce [34] associates the invariant, J α (σ ) ∈ Q, as a counting
invariant of semistable objects E with [E] = α. The definition of J α (σ ) is
similar to the virtual Euler number of the moduli space of σ -semistable objects
of type α. It is known that there is a nice moduli space of semistable A-
modules, so its virtual Euler number is well-defined. (See [41].) However for
our purpose, it doesn’t work well, when there is a σ -semistable object which
is not stable. Actually we have to take account of the automorphism groups of
semistable objects, so Joyce works in the context of moduli stacks of semistable
objects, which are known to be algebraic stacks. Roughly speaking J α (σ ) is a
“weighted Euler number” of the moduli stack of σ -semistable objects of type
α, and especially when α is primitive and σ is generic, the invariant J α (σ )
coincides with the virtual Euler number of its coarse moduli space. For the
detail, see [34]. We set
α (σ ) as follows,
α (σ ) = J α (σ )cα ∈ Lα .
semistable objects does not change in a chamber. (See [15].) This implies that
σ
→
α (σ ) is constant in a chamber, however it will be a different value under a
wall crossing. The idea of [33] is to cancel out the discontinuity of σ
→
α (σ ),
by using the (also discontinuous) functions,
Fn : (C∗ )n −→ C,
and make a function f α : Stab(A) → U (L),
f α (σ ) = Fn (Z(α1 ), · · · , Z(αn ))
α1 (σ ) ∗ · · · ∗
αn (σ ). (18)
α1 +···+αn =α
2π i 1
2 log
(2πi)
z2
, z1
z 2
∈ (0, ∞).
z1
where β, γ ∈ C(A).
Explicitly the function f α is expanded in the following formula,
n
f α (σ ) = Fn (Z(α1 ), · · · , Z(σn )) J αi (σ )
α1 +···+αn =α i=1
⎡ ⎤
1
×⎣ χ (αi , αj )⎦ ,
2n−1 i→j in
with αi ∈ C(A), is a connected and simply connected oriented graph with
vertex {1, · · · , n}, and if there is an arrow i → j in , we must have i ≤ j .
448 Yukinobu Toda
Acknowledgement
The author thanks Hokuto Uehara for checking the manuscript and giving the
nice advice. He is supported by Japan Society for the Promotion of Sciences
Research Fellowships for Young Scientists, No 198007.
References
[1] P. Aspinwall. A Point’s Point of View of Stringy Geometry. J. High Energy Phys,
Vol. 002, p. 15pp, 2003.
[2] A. Beilinson. Coherent sheaves on Pn and problems of linear algebra. Funct. Anal.
Appl, Vol. 12, pp. 214–216, 1978.
[3] A. Beilinson, J. Bernstein, and P. Deligne. Faisceaux pervers. Analysis and topol-
ogy on singular spaces I, Asterisque, Vol. 100, pp. 5–171, 1982.
[4] O. Ben-Bassat, J. Block, and T. Pantev. Non-commutative tori and Fourier-Mukai
duality. Compositio Math, Vol. 143, pp. 423–475, 2007.
[5] A. Bergman. Stability conditions and Branes at Singularities. preprint.
math.AG/0702092.
[6] R. Bezrukavnikov and D. Kaledin. McKay equivalence for symplectic resolutions
of quotient singularities. Proc. Steklov Inst. Math, Vol. 246, pp. 13–33, 2004.
[7] A. Bondal and D. Orlov. Semiorthgonal decomposition for algebraic varieties.
preprint. math.AG/9506012.
[8] A. Bondal and D. Orlov. Reconstruction of a variety from the derived category
and groups of autoequivalences. Compositio Math, Vol. 125, pp. 327–344, 2001.
[9] T. Bridgeland. Spaces of stability conditions. preprint. math.AG/0611510.
[10] T. Bridgeland. Stability conditions and Kleinian singularities. preprint.
math.AG/0508257.
[11] T. Bridgeland. Flops and derived categories. Invent. Math, Vol. 147, pp. 613–632,
2002.
[12] T. Bridgeland. Derived categories of coherent sheaves. Proceedings of the 2006
ICM, 2006. math.AG/0602129.
[13] T. Bridgeland. Stability conditions on a non-compact Calabi-Yau threefold.
Comm. Math. Phys., Vol. 266, pp. 715–733, 2006.
[14] T. Bridgeland. Stability conditions on triangulated categories. Ann. of Math,
Vol. 166, pp. 317–345, 2007.
[15] T. Bridgeland. Stability conditions on K3 surfaces. Duke Math. J., Vol. 141,
pp. 241–291, 2008.
Derived categories of coherent sheaves on algebraic varieties 449
0. Introduction
This short article is based on a lecture I gave at the “Workshop on Triangulated
Categories”, Leeds, August 2006. It is a survey of recent results on rigid
dualizing complexes over commutative rings. Most of these results are joint
work of mine with James Zhang. The idea of rigid dualizing complex is due to
Michel Van den Bergh.
By default all rings considered in this article are commutative. We begin by
recalling the notion of dualizing complex over a noetherian ring A. Next let B
be a noetherian A-algebra. We define what is a rigid complex of B-modules
relative to A. In making this definition we must use differential graded algebras
(when B is not flat over A). The functorial properties of rigid complexes are
explained. We then discuss rigid dualizing complexes, which by definition
are complexes that are both rigid and dualizing. Finally we show how rigid
complexes can be used to understand Cohen-Macaulay homomorphisms and
relative dualizing sheaves.
I wish to thank my collaborator James Zhang. Thanks also to Luchezar
Avramov, Srikanth Iyengar and Joseph Lipman for discussions regarding the
material in Section 5.
Key words and phrases. commutative rings, DG algebras, derived categories, rigid complexes.
Mathematics Subject Classification 2000. Primary: 18E30; Secondary: 18G10, 16E45, 18G15.
This research was supported by the US-Israel Binational Science Foundation.
452
Rigid dualizing complexes 453
Condition (i) means that there is an integer d such that ExtiA (M, R) = 0 for
all i > d and all modules M.
Recall that a noetherian ring K is called regular if all its local rings Kp ,
p ∈ Spec K, are regular local rings.
R := K ∈ Dbf (Mod K)
in D(Mod B).
Now suppose A is any ring. Trying to write A instead of K in formula (2.1)
does not make sense: instead of M ⊗A M we must take the derived tensor
product M ⊗LA M; but then there is no obvious way to make M ⊗LA M into a
complex of B ⊗A B -modules.
The problem is torsion: B might fail to be a flat A-algebra. This is where
differential graded algebras (DG algebras) enter the picture.
A DG algebra is a graded ring à = i∈Z Ãi , together with a graded deriva-
tion d : Ã → Ã of degree 1, satisfying d ◦ d = 0.
A DG algebra quasi-isomorphism is a homomorphism f : Ã → B̃ respect-
ing degrees, multiplications and differentials, and such that H(f ) : HÃ → HB̃
is an isomorphism (of graded algebras).
We shall only consider super-commutative non-positive DG algebras. Super-
commutative means that ab = (−1)ij ba and c2 = 0 for all a ∈ Ãi , b ∈ Ãj and
c ∈ Ã2i+1 . Non-positive means that à = i≤0 Ãi .
We view a ring A as a DG algebra concentrated in degree 0. Given a DG
algebra homomorphism A → à we say that à is a DG A-algebra.
Let A be a ring. A semi-free DG A-algebra is a DG A-algebra Ã, such
that after forgetting the differential à is isomorphic, as graded A-algebra, to a
super-polynomial algebra on some graded set of variables.
is an equivalence.
Getting back to our original problem, suppose A is a ring and B is an A-
algebra. Choose a semi-free DG algebra resolution B̃ → B relative to A. For
M ∈ D(Mod B) define
in D(Mod B).
The functor SqB/A , called the squaring operation, is nonlinear. In fact, given
a morphism φ : M → M in D(Mod B) and an element b ∈ B one has
in
is an isomorphism in D(Mod B). Then the pair (M, ρ) is called a rigid complex
over B relative to A.
Definition 2.7. Say (M, ρ) and (N, σ ) are rigid complexes over B relative to
A. A morphism φ : M → N in D(Mod B) is called a rigid morphism relative
456 Amnon Yekutieli
to A if the diagram
ρ
M −−−−→ SqB/A M
⏐ ⏐
⏐
φ
⏐Sq (φ)
B/A
σ
N −−−−→ SqB/A N
is commutative.
We denote by Dbf (Mod B)rig/A the category of rigid complexes over B relative
to A.
B → HomD(Mod B) (M, M)
is bijective. Then the only automorphism of (M, ρ) in Dbf (Mod B)rig/A is the
identity 1M .
f : Spec B → Spec A.
Rigid dualizing complexes 457
The assignment
(M, ρ)
→ f & (M, ρ) := f & (ρ), f & M
is functorial.
(1) The algebra A has a rigid dualizing complex (RA , ρA ), which is unique up
to a unique rigid isomorphism.
(2) Given a finite homomorphism f ∗ : A → B, there is a unique rigid isomor-
phism f (RA , ρA ) → (RB , ρB ).
(3) Given an essentially smooth homomorphism f ∗ : A → B , there is a unique
rigid isomorphism f & (RA , ρA ) → (RB , ρB ).
which is dualizing (cf. Examples 1.2 and 2.8). Now the structural homomor-
phism K → A can be factored into
f∗ g∗ h∗
K−
→B−
→C−
→ A,
by the formula
f ! M := RHomB B ⊗LA RHomA (M, RA ), RB .
It is easy to show that the assignment f ∗
→ f ! is a pseudofunctor from
the category EFTAlg /K to the 2-category Cat of all categories. Moreover,
using Theorem 4.2 one can show that this operation has very good properties.
For instance, when f ∗ is finite, then there is a functorial nondegenerate trace
morphism
Trf : f ! M → M.
Remark 4.4. According to Grothendieck’s duality theory in [RD], if f : X →
Y is a finite type morphism between noetherian schemes, and if Y has a dualizing
complex, then there is a functor
f !(G) : D+ +
c (Mod OY ) → Dc (Mod OX ),
in D(Mod B).
Condition (*) implies that the support of the complex RB/A is Spec B. One
can prove that
f !M ∼
= RB/A ⊗LA M
Corollary 5.3. Assume that in Theorem 5.2 the ring A is Gorenstein. Then
RB/A is a dualizing complex over B
in D(Mod B).
The module
ωB/A := ωBi /A
i
B ∼
= A ⊗A B,
ωB /A ∼
= A ⊗A ωB/A
Remark 5.7. The recent paper [AI] contains results similar to Theorem 5.4
and Corollary 5.6, obtained by different methods, and without the requirement
that A is tractable.
References
[AK] A. Altman and S. Kleiman, “Introduction to Grothendieck Duality,” Lecture
Notes in Math. 20, Springer, 1970.
[AI] L.L. Avramov and S. Iyengar, Gorenstein algebras and Hochschild cohomol-
ogy, Michigan Math. J. 57 (2008), 17–35.
[AJL] L. Alonso, A. Jeremı́as and J. Lipman, Duality and flat base change on formal
schemes, in “Studies in Duality on Noetherian Formal Schemes and Non-
Noetherian Ordinary Schemes,” Contemp. Math. 244, Amer. Math. Soc., 1999,
3–90.
[Be] K. Behrend, Differential Graded Schemes I: Perfect Resolving Algebras, eprint
arXiv:math/0212225v1 [math.AG] at http://arXiv.org.
[Co] B. Conrad, “Grothendieck Duality and Base Change,” Lecture Notes in Math.
1750, Springer, 2000.
[Hi] V. Hinich, Homological algebra of homotopy algebras, Comm. Algebra 25
(1997), no. 10, 3291–3323.
[HS] R. Hübl and P. Sastry, Regular differential forms and relative duality, Amer. J.
Math. 115 (1993), no. 4, 749–787.
[HK] R. Hübl and E. Kunz, Regular differential forms and duality for projective
morphisms, J. Reine Angew. Math. 410 (1990), 84–108.
[Hu] R. Hübl, “Traces of Differential Forms and Hochschild Homology,” Lecture
Notes in Math. 1368, Springer, 1989.
[Ke] B. Keller, Deriving DG categories, Ann. Sci. École Norm. Sup. (4) 27 (1994),
no. 1, 63–102.
[Li] J. Lipman, “Residues and Traces of Differential Forms via Hochschild Homol-
ogy,” Contemporary Mathematics 61, Amer. Math. Soc., Providence, RI, 1987.
[Ne] A. Neeman, The Grothendieck duality theorem via Bousfield’s techniques and
Brown representability, J. Amer. Math. Soc. 9 (1996), no. 1, 205–236.
[RD] R. Hartshorne, “Residues and Duality,” Lecture Notes in Math. 20, Springer-
Verlag, Berlin, 1966.
Rigid dualizing complexes 463
[VdB] M. Van den Bergh, Existence theorems for dualizing complexes over non-
commutative graded and filtered ring, J. Algebra 195 (1997), no. 2, 662–679.
[Ye1] A. Yekutieli, “An Explicit Construction of the Grothendieck Residue Complex”
(with an appendix by P. Sastry), Astérisque 208 (1992).
[Ye2] A. Yekutieli, Rigidity, Residues, and Grothendieck Duality for Schemes, in
preparation.
[YZ1] A. Yekutieli and J.J. Zhang, Rings with Auslander dualizing complexes, J.
Algebra 213 (1999), no. 1, 1–51.
[YZ2] A. Yekutieli and J.J. Zhang, Rigid Dualizing Complexes and Perverse Sheaves
over Differential Algebras, Compositio Math. 141 (2005), 620–654.
[YZ3] A. Yekutieli and J.J. Zhang, Dualizing Complexes and Perverse Sheaves on
Noncommutative Ringed Schemes, Selecta Math. 12 (2006), 137–177.
[YZ4] A. Yekutieli and J.J. Zhang, Rigid Complexes via DG Algebras, Trans. AMS
360 no. 6 (2008), 3211–3248.
[YZ5] A. Yekutieli and J.J. Zhang, Rigid Dualizing Complexes over Commutative
Rings, Algebr. Represent. Theory, 12, no. 1 (2009), 19–52.